From 38dd62819b3788a5340dbb70ed1423a0bbca1e44 Mon Sep 17 00:00:00 2001
From: Diego Rodrigues de Sa e Souza
 <8016841+diegosouzapw@users.noreply.github.com>
Date: Sat, 18 Jul 2026 03:16:58 -0300
Subject: [PATCH] feat(providers): Speechmatics STT, gTTS, VibeProxy preset
 (#6659, #6667, #6874) (#7655)

Validated in merge-train --fast @ 6cafcbb (static gates + 9 changed test files + vitest green, 2m35s; full suite ran today on train 2c tip)
---
 .../6659-speechmatics-stt-provider.md         |   1 +
 .../features/6667-gtts-audio-tts-provider.md  |   1 +
 ...4-vibeproxy-openai-provider-node-preset.md |   1 +
 docs/reference/PROVIDER_REFERENCE.md          |   9 +-
 open-sse/config/audioRegistry.ts              |  29 +++
 open-sse/executors/gtts.ts                    | 214 ++++++++++++++++++
 open-sse/handlers/audioSpeech.ts              |  29 ++-
 open-sse/handlers/audioTranscription.ts       | 104 ++++++++-
 src/app/api/provider-nodes/route.ts           |  60 ++++-
 src/shared/constants/providers/audio.ts       |  12 +
 src/shared/validation/schemas/provider.ts     |  38 +++-
 tests/unit/gtts-provider.test.ts              | 187 +++++++++++++++
 .../provider-nodes-vibeproxy-preset.test.ts   | 135 +++++++++++
 ...peechmatics-transcription-provider.test.ts | 143 ++++++++++++
 14 files changed, 951 insertions(+), 12 deletions(-)
 create mode 100644 changelog.d/features/6659-speechmatics-stt-provider.md
 create mode 100644 changelog.d/features/6667-gtts-audio-tts-provider.md
 create mode 100644 changelog.d/features/6874-vibeproxy-openai-provider-node-preset.md
 create mode 100644 open-sse/executors/gtts.ts
 create mode 100644 tests/unit/gtts-provider.test.ts
 create mode 100644 tests/unit/provider-nodes-vibeproxy-preset.test.ts
 create mode 100644 tests/unit/speechmatics-transcription-provider.test.ts

diff --git a/changelog.d/features/6659-speechmatics-stt-provider.md b/changelog.d/features/6659-speechmatics-stt-provider.md
new file mode 100644
index 0000000000..e19ecb31b8
--- /dev/null
+++ b/changelog.d/features/6659-speechmatics-stt-provider.md
@@ -0,0 +1 @@
+- **feat(providers):** add Speechmatics as an STT provider — async batch transcription (Enhanced operating point), 8 hours/month free tier, no credit card required. Streaming (real-time) mode is out of scope for v1. (#6659)
diff --git a/changelog.d/features/6667-gtts-audio-tts-provider.md b/changelog.d/features/6667-gtts-audio-tts-provider.md
new file mode 100644
index 0000000000..3c55e36deb
--- /dev/null
+++ b/changelog.d/features/6667-gtts-audio-tts-provider.md
@@ -0,0 +1 @@
+- **feat(providers):** add gTTS (Google Translate TTS) as a free, no-signup audio-speech provider — routes through Google's current `batchexecute` RPC endpoint (the previously proposed `translate_tts` endpoint is deprecated), splitting input at the 100-char-per-request limit. (#6667)
diff --git a/changelog.d/features/6874-vibeproxy-openai-provider-node-preset.md b/changelog.d/features/6874-vibeproxy-openai-provider-node-preset.md
new file mode 100644
index 0000000000..8a55e40627
--- /dev/null
+++ b/changelog.d/features/6874-vibeproxy-openai-provider-node-preset.md
@@ -0,0 +1 @@
+- **feat(providers):** add a `vibeproxy-openai` provider-node preset to `POST /api/provider-nodes` — defaults name/prefix/apiType for VibeProxy's local OpenAI-compatible gateway and normalizes the caller-supplied base URL to its `/v1` root; `baseUrl` remains mandatory. (#6874, idea from #6137 by @KooshaPari)
diff --git a/docs/reference/PROVIDER_REFERENCE.md b/docs/reference/PROVIDER_REFERENCE.md
index 74298874cc..73532665cf 100644
--- a/docs/reference/PROVIDER_REFERENCE.md
+++ b/docs/reference/PROVIDER_REFERENCE.md
@@ -1,16 +1,16 @@
 ---
 title: "Provider Reference"
 version: 3.8.49
-lastUpdated: 2026-07-17
+lastUpdated: 2026-07-18
 ---
 
 # Provider Reference
 
 > **Auto-generated** from `src/shared/constants/providers.ts` — do not edit by hand.
 > Regenerate with: `npm run gen:provider-reference`
-> **Last generated:** 2026-07-17
+> **Last generated:** 2026-07-18
 
-Total providers: **264**. See category breakdown below.
+Total providers: **265**. See category breakdown below.
 
 ## Categories
 
@@ -305,7 +305,7 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each
 | `tavily-search` | `tavily-search` | Tavily Search | Search | [link](https://tavily.com) | API key from app.tavily.com (format: tvly-...) |
 | `youcom-search` | `youcom-search` | You.com Search | Search | [link](https://you.com/business/api/) | X-API-Key from the You.com platform dashboard |
 
-## Audio-only Providers (9)
+## Audio-only Providers (10)
 
 | ID | Alias | Name | Tags | Website | Notes |
 |----|-------|------|------|---------|-------|
@@ -318,6 +318,7 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each
 | `inworld` | `inworld` | Inworld | Audio | [link](https://inworld.ai) | — |
 | `playht` | `playht` | PlayHT | Audio | [link](https://play.ht) | — |
 | `rev-ai` | `revai` | Rev AI | Audio | [link](https://www.rev.ai) | — |
+| `speechmatics` | `sm` | Speechmatics | Audio | [link](https://www.speechmatics.com) | Free tier — 8 hours/month, no credit card required. Batch (async) mode only. |
 
 ## Upstream Proxy Providers (2)
 
diff --git a/open-sse/config/audioRegistry.ts b/open-sse/config/audioRegistry.ts
index 117e9a4b52..3f56c66ec8 100644
--- a/open-sse/config/audioRegistry.ts
+++ b/open-sse/config/audioRegistry.ts
@@ -189,6 +189,21 @@ export const AUDIO_TRANSCRIPTION_PROVIDERS: Record<string, AudioProvider> = {
       { id: "fusion", name: "Fusion ASR" },
     ],
   },
+
+  speechmatics: {
+    id: "speechmatics",
+    // POST https://asr.api.speechmatics.com/v2/jobs — async batch workflow:
+    // submit multipart job (audio + JSON config) → poll → fetch transcript.
+    // Auth: Authorization: Bearer <api-key>
+    // Free tier: 8 hours/month, no credit card required.
+    // Streaming (WebSocket real-time) mode is out of scope for v1 — batch only.
+    baseUrl: "https://asr.api.speechmatics.com/v2/jobs",
+    authType: "apikey",
+    authHeader: "bearer",
+    async: true,
+    format: "speechmatics",
+    models: [{ id: "enhanced", name: "Enhanced" }],
+  },
 };
 
 /**
@@ -460,6 +475,20 @@ export const AUDIO_SPEECH_PROVIDERS: Record<string, AudioProvider> = {
     ],
   },
 
+  gtts: {
+    id: "gtts",
+    // Google Translate TTS — reverse-engineered, no API key required.
+    // POST batchexecute RPC (unlike the deprecated GET /translate_tts) —
+    // handled by open-sse/executors/gtts.ts, dispatched via the "gtts" format.
+    // No official SLA; per-IP rate-limited by Google without notice.
+    baseUrl: "https://translate.google.com/_/TranslateWebserverUi/data/batchexecute",
+    authType: "none",
+    authHeader: "none",
+    format: "gtts",
+    supportedFormats: ["mp3"],
+    models: [{ id: "default", name: "Google Translate TTS (Free)" }],
+  },
+
   "xiaomi-mimo": {
     id: "xiaomi-mimo",
     baseUrl: "https://api.xiaomimimo.com/v1/chat/completions",
diff --git a/open-sse/executors/gtts.ts b/open-sse/executors/gtts.ts
new file mode 100644
index 0000000000..a72c44b2f7
--- /dev/null
+++ b/open-sse/executors/gtts.ts
@@ -0,0 +1,214 @@
+/**
+ * gTTS — Google Translate text-to-speech (#6667).
+ *
+ * Reverse-engineered, unofficial, undocumented endpoint (not a published
+ * Google public API) — the same class of integration this codebase already
+ * accepts for other "-web"/no-auth style providers (edgeTts.ts, chipotle.ts).
+ * No user account/API key is required.
+ *
+ * The issue's originally proposed endpoint
+ * (`https://translate.google.com/translate_tts`, GET with `q`/`tl`/`ie` query
+ * params) has been deprecated by Google. The current, working mechanism —
+ * verified directly against `pndurette/gTTS`'s `gtts/tts.py` source — is a
+ * POST RPC call to Google's internal `batchexecute` endpoint:
+ *
+ *   POST https://translate.google.<tld>/_/TranslateWebserverUi/data/batchexecute
+ *   Content-Type: application/x-www-form-urlencoded;charset=utf-8
+ *   Body: f.req=<urlencoded JSON envelope>
+ *
+ * The envelope wraps `[text, lang, true, "null"]` under RPC id `"jQ1olc"`:
+ *   f.req = [[["jQ1olc", '["<text>","<lang>",true,"null"]', null, "generic"]]]
+ *
+ * There is a hard 100-character-per-request limit on `text` — longer input
+ * must be split into multiple RPC calls and the resulting MP3 byte chunks
+ * concatenated (§ `chunkGttsText`).
+ *
+ * The response is a `)]}'`-prefixed "batchexecute" payload; the base64 audio
+ * lives inside the entry whose outer array starts with `["wrb.fr","jQ1olc",…]`
+ * (§ `parseBatchExecuteResponse`).
+ *
+ * All parsing/chunking above is implemented as pure functions so it can be
+ * unit-tested without a live upstream connection — only `synthesizeGtts()`
+ * itself touches the network, and it accepts an injectable `fetch` for tests.
+ */
+
+/** Hard per-request character limit enforced by Google's batchexecute endpoint. */
+export const GOOGLE_TTS_MAX_CHARS = 100;
+
+const GTTS_RPC_ID = "jQ1olc";
+const GTTS_USER_AGENT =
+  "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36";
+const GTTS_REFERER = "http://translate.google.com/";
+const DEFAULT_LANG = "en";
+const DEFAULT_TLD = "com";
+/** Only allow simple BCP-47-ish language codes to keep this untrusted input from injecting RPC payload structure. */
+const LANG_PATTERN = /^[a-z]{2,3}(-[A-Za-z0-9]{2,8})?$/;
+
+export class GttsUpstreamError extends Error {
+  status: number;
+  constructor(status: number, message: string) {
+    super(message);
+    this.name = "GttsUpstreamError";
+    this.status = status;
+  }
+}
+
+export interface GttsSynthInput {
+  text: string;
+  lang?: string;
+  tld?: string;
+}
+
+/** Normalize a caller-supplied language code, falling back to English. */
+export function normalizeGttsLang(lang: unknown): string {
+  const value = typeof lang === "string" ? lang.trim() : "";
+  return LANG_PATTERN.test(value) ? value : DEFAULT_LANG;
+}
+
+/**
+ * Split `text` into chunks respecting Google's 100-character-per-request
+ * limit, preferring to break on whitespace so words are not split mid-token.
+ * A single "word" longer than `maxChars` is hard-split as a last resort.
+ */
+export function chunkGttsText(text: unknown, maxChars: number = GOOGLE_TTS_MAX_CHARS): string[] {
+  const trimmed = typeof text === "string" ? text.trim() : "";
+  if (!trimmed) return [];
+  if (trimmed.length <= maxChars) return [trimmed];
+
+  const chunks: string[] = [];
+  let remaining = trimmed;
+  while (remaining.length > maxChars) {
+    let splitAt = -1;
+    for (let i = maxChars; i > 0; i--) {
+      if (/\s/.test(remaining[i])) {
+        splitAt = i;
+        break;
+      }
+    }
+    if (splitAt <= 0) splitAt = maxChars;
+    chunks.push(remaining.slice(0, splitAt).trim());
+    remaining = remaining.slice(splitAt).trim();
+  }
+  if (remaining) chunks.push(remaining);
+  return chunks.filter((c) => c.length > 0);
+}
+
+/** Build the `f.req=`-prefixed, urlencoded RPC body for one text chunk. */
+export function buildGttsRpcBody(text: string, lang: string): string {
+  const innerPayload = JSON.stringify([text, lang, true, "null"]);
+  const envelope = [[[GTTS_RPC_ID, innerPayload, null, "generic"]]];
+  return `f.req=${encodeURIComponent(JSON.stringify(envelope))}&`;
+}
+
+/**
+ * Extract the base64 audio payload from one `["wrb.fr","jQ1olc",…]` entry,
+ * or `null` if this entry isn't a matching audio fragment.
+ */
+function extractAudioFromWrbFrEntry(entry: unknown): string | null {
+  if (!Array.isArray(entry) || entry[0] !== "wrb.fr" || entry[1] !== GTTS_RPC_ID) return null;
+  if (typeof entry[2] !== "string") return null;
+
+  try {
+    const inner = JSON.parse(entry[2]);
+    if (Array.isArray(inner) && typeof inner[0] === "string" && inner[0].length > 0) {
+      return inner[0];
+    }
+  } catch {
+    // Not a JSON-parseable payload — treat as "no audio in this entry".
+  }
+  return null;
+}
+
+/** Parse one newline-delimited JSON fragment, returning its audio payload if present. */
+function findAudioInBatchExecuteLine(line: string): string | null {
+  let outer: unknown;
+  try {
+    outer = JSON.parse(line);
+  } catch {
+    return null;
+  }
+  if (!Array.isArray(outer)) return null;
+
+  for (const entry of outer) {
+    const audio = extractAudioFromWrbFrEntry(entry);
+    if (audio) return audio;
+  }
+  return null;
+}
+
+/**
+ * Extract the base64-encoded audio payload from a `batchexecute` response.
+ * The response is `)]}'`-prefixed, followed by newline-delimited JSON
+ * fragments interleaved with numeric length-prefix lines; the audio lives
+ * in the fragment whose entry starts with `["wrb.fr","jQ1olc",…]`.
+ */
+export function parseBatchExecuteResponse(raw: string): string {
+  const cleaned = typeof raw === "string" ? raw.replace(/^\)\]\}'\n?/, "") : "";
+  const lines = cleaned.split("\n").filter((line) => {
+    const trimmedLine = line.trim();
+    return trimmedLine.length > 0 && !/^\d+$/.test(trimmedLine);
+  });
+
+  for (const line of lines) {
+    const audio = findAudioInBatchExecuteLine(line);
+    if (audio) return audio;
+  }
+
+  throw new GttsUpstreamError(502, "gTTS response did not contain audio data");
+}
+
+type FetchLike = (url: string, init: RequestInit) => Promise<Response>;
+
+/** Synthesize one ≤100-char chunk, returning the decoded MP3 bytes. */
+async function synthesizeGttsChunk(
+  chunk: string,
+  lang: string,
+  tld: string,
+  fetchImpl: FetchLike
+): Promise<Buffer> {
+  const body = buildGttsRpcBody(chunk, lang);
+  const res = await fetchImpl(
+    `https://translate.google.${tld}/_/TranslateWebserverUi/data/batchexecute`,
+    {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/x-www-form-urlencoded;charset=utf-8",
+        "User-Agent": GTTS_USER_AGENT,
+        Referer: GTTS_REFERER,
+      },
+      body,
+    }
+  );
+
+  if (!res.ok) {
+    const errText = await res.text().catch(() => "");
+    throw new GttsUpstreamError(res.status, errText || `gTTS upstream error (${res.status})`);
+  }
+
+  const raw = await res.text();
+  const base64Audio = parseBatchExecuteResponse(raw);
+  return Buffer.from(base64Audio, "base64");
+}
+
+/**
+ * Synthesize `input.text` end-to-end: chunk to Google's 100-char limit,
+ * POST each chunk to the batchexecute RPC endpoint, and concatenate the
+ * decoded MP3 byte chunks into one buffer.
+ */
+export async function synthesizeGtts(
+  input: GttsSynthInput,
+  fetchImpl: FetchLike = fetch
+): Promise<Buffer> {
+  const lang = normalizeGttsLang(input.lang);
+  const tld = (typeof input.tld === "string" && input.tld.trim()) || DEFAULT_TLD;
+  const chunks = chunkGttsText(input.text);
+  if (chunks.length === 0) {
+    throw new GttsUpstreamError(400, "gTTS requires non-empty input text");
+  }
+
+  const buffers: Buffer[] = [];
+  for (const chunk of chunks) {
+    buffers.push(await synthesizeGttsChunk(chunk, lang, tld, fetchImpl));
+  }
+  return Buffer.concat(buffers);
+}
diff --git a/open-sse/handlers/audioSpeech.ts b/open-sse/handlers/audioSpeech.ts
index d7053a1b09..f885a59e69 100644
--- a/open-sse/handlers/audioSpeech.ts
+++ b/open-sse/handlers/audioSpeech.ts
@@ -23,6 +23,7 @@ import { kieExecutor } from "../executors/kie.ts";
 import { vertexGenerateSpeech } from "../executors/vertexMedia.ts";
 import { handleAwsPollySpeech } from "../executors/awsPollyTts.ts";
 import { handleEdgeTtsSpeech } from "../executors/edgeTts.ts";
+import { GttsUpstreamError, normalizeGttsLang, synthesizeGtts } from "../executors/gtts.ts";
 import { errorResponse } from "../utils/error.ts";
 import { audioStreamResponse, upstreamErrorResponse } from "../utils/audioResponse.ts";
 import {
@@ -725,6 +726,28 @@ async function handleTortoiseSpeech(providerConfig, body) {
   });
 }
 
+/**
+ * Handle gTTS TTS (local no-auth, Google Translate batchexecute RPC).
+ * `voice` doubles as the language code since gTTS has no voice concept —
+ * defaults to English when omitted or unrecognized.
+ */
+async function handleGttsSpeech(body) {
+  try {
+    const audio = await synthesizeGtts({
+      text: body.input,
+      lang: normalizeGttsLang(body.voice),
+    });
+    return new Response(audio, {
+      status: 200,
+      headers: { ...CORS_HEADERS, "Content-Type": "audio/mpeg" },
+    });
+  } catch (err) {
+    const status = err instanceof GttsUpstreamError ? err.status : 502;
+    const message = err instanceof Error ? err.message : "gTTS synthesis failed";
+    return errorResponse(status, message);
+  }
+}
+
 /**
  * Handle audio speech (TTS) request
  *
@@ -761,7 +784,7 @@ export async function handleAudioSpeech({
   if (!providerConfig) {
     return errorResponse(
       400,
-      `No speech provider found for model "${body.model}". Use format provider/model. Available: openai, hyperbolic, deepgram, nvidia, elevenlabs, huggingface, inworld, cartesia, playht, kie, aws-polly, xiaomi-mimo, edgetts, coqui, tortoise, qwen`
+      `No speech provider found for model "${body.model}". Use format provider/model. Available: openai, hyperbolic, deepgram, nvidia, elevenlabs, huggingface, inworld, cartesia, playht, kie, aws-polly, xiaomi-mimo, edgetts, gtts, coqui, tortoise, qwen`
     );
   }
 
@@ -830,6 +853,10 @@ export async function handleAudioSpeech({
       return handleEdgeTtsSpeech(body, clientIp);
     }
 
+    if (providerConfig.format === "gtts") {
+      return handleGttsSpeech(body);
+    }
+
     if (providerConfig.format === "xiaomi-mimo-tts") {
       return handleXiaomiMimoSpeech(providerConfig, body, modelId, token, credentials);
     }
diff --git a/open-sse/handlers/audioTranscription.ts b/open-sse/handlers/audioTranscription.ts
index 703004b21d..692b30df35 100644
--- a/open-sse/handlers/audioTranscription.ts
+++ b/open-sse/handlers/audioTranscription.ts
@@ -531,6 +531,104 @@ async function handleRevAiTranscription(providerConfig, file, modelId, token) {
   return errorResponse(504, "Rev AI transcription timed out after 120s");
 }
 
+/**
+ * Speechmatics operating point (accuracy tier). Catalog model ids are the
+ * real Speechmatics `operating_point` values ("standard", "enhanced",
+ * "melia-1"), so this passes straight through — kept as a named seam in
+ * case a future catalog id needs remapping.
+ */
+function speechmaticsOperatingPoint(modelId: string): string {
+  return modelId;
+}
+
+/**
+ * Fetch and return the finished Speechmatics transcript once a job reaches
+ * the "done" state.
+ */
+async function fetchSpeechmaticsTranscript(jobUrl, authHeaders) {
+  const transcriptRes = await fetch(`${jobUrl}/transcript?format=txt`, {
+    headers: { ...authHeaders, Accept: "text/plain" },
+  });
+  if (!transcriptRes.ok) {
+    return upstreamErrorResponse(transcriptRes, await transcriptRes.text());
+  }
+  const text = await transcriptRes.text();
+  return Response.json({ text: text || "" }, { headers: { ...CORS_HEADERS } });
+}
+
+function speechmaticsJobErrorMessage(result): string {
+  const errors = result?.job?.errors;
+  const first = Array.isArray(errors) ? errors[0] : null;
+  return first?.message || "Speechmatics transcription failed";
+}
+
+/**
+ * Poll a submitted Speechmatics job until it reaches a terminal state
+ * (max 120s), then fetch its transcript.
+ */
+async function pollSpeechmaticsJob(jobUrl, authHeaders) {
+  const maxWait = 120_000;
+  const start = Date.now();
+
+  while (Date.now() - start < maxWait) {
+    await new Promise((r) => setTimeout(r, 2000));
+
+    const pollRes = await fetch(jobUrl, { headers: authHeaders });
+    if (!pollRes.ok) continue;
+
+    const result = await pollRes.json();
+    const status = result?.job?.status;
+
+    if (status === "done") {
+      return fetchSpeechmaticsTranscript(jobUrl, authHeaders);
+    }
+
+    if (status === "rejected") {
+      return errorResponse(500, speechmaticsJobErrorMessage(result));
+    }
+  }
+
+  return errorResponse(504, "Speechmatics transcription timed out after 120s");
+}
+
+/**
+ * Handle Speechmatics transcription (async batch: submit multipart job → poll → fetch transcript)
+ *
+ * Speechmatics batch mode accepts the audio file directly in the job-submission
+ * multipart body (field "data_file") alongside a JSON "config" field describing
+ * the requested transcription options. Streaming (real-time WebSocket) mode is
+ * out of scope for v1 — this handler only implements batch (REST) transcription.
+ */
+async function handleSpeechmaticsTranscription(providerConfig, file, modelId, token) {
+  const authHeaders = buildAuthHeaders(providerConfig, token);
+  const baseUrl = providerConfig.baseUrl.replace(/\/$/, "");
+
+  // Step 1: submit the job — multipart body with "data_file" (audio) + "config" (JSON)
+  const config = JSON.stringify({
+    type: "transcription",
+    transcription_config: { operating_point: speechmaticsOperatingPoint(modelId) },
+  });
+  const { body, contentType } = await buildMultipartBody(file, { config }, "data_file");
+
+  const submitRes = await fetch(baseUrl, {
+    method: "POST",
+    headers: { ...authHeaders, "Content-Type": contentType },
+    body,
+  });
+
+  if (!submitRes.ok) {
+    return upstreamErrorResponse(submitRes, await submitRes.text());
+  }
+
+  const { id: jobId } = await submitRes.json();
+  if (!jobId) {
+    return errorResponse(502, "Speechmatics did not return a job id");
+  }
+
+  // Step 2: poll for completion (max 120s)
+  return pollSpeechmaticsJob(`${baseUrl}/${jobId}`, authHeaders);
+}
+
 /**
  * Handle audio transcription request
  *
@@ -573,7 +671,7 @@ export async function handleAudioTranscription({
   if (!providerConfig) {
     return errorResponse(
       400,
-      `No transcription provider found for model "${model}". Available: openai, groq, deepgram, assemblyai, nvidia, huggingface, qwen, gladia, rev-ai`
+      `No transcription provider found for model "${model}". Available: openai, groq, deepgram, assemblyai, nvidia, huggingface, qwen, gladia, rev-ai, speechmatics`
     );
   }
 
@@ -639,6 +737,10 @@ export async function handleAudioTranscription({
     return handleRevAiTranscription(providerConfig, file, modelId, token);
   }
 
+  if (providerConfig.format === "speechmatics") {
+    return handleSpeechmaticsTranscription(providerConfig, file, modelId, token);
+  }
+
   // Default: OpenAI/Groq/Qwen3-compatible multipart proxy
   const extraFields: Record<string, string> = {};
   for (const key of [
diff --git a/src/app/api/provider-nodes/route.ts b/src/app/api/provider-nodes/route.ts
index 87bb5330b2..e6dbeda5cf 100644
--- a/src/app/api/provider-nodes/route.ts
+++ b/src/app/api/provider-nodes/route.ts
@@ -19,6 +19,34 @@ const ANTHROPIC_COMPATIBLE_DEFAULTS = {
   baseUrl: "https://api.anthropic.com/v1",
 };
 
+// #6874: VibeProxy (github.com/automazeio/vibeproxy) — local OpenAI-compatible
+// gateway. baseUrl has no default (operator-specific host/port); name/prefix/
+// apiType are filled in when the caller omits them.
+const VIBEPROXY_OPENAI_DEFAULTS = {
+  name: "VibeProxy",
+  prefix: "vibeproxy",
+  apiType: "chat" as const,
+};
+
+/**
+ * Normalize a caller-supplied VibeProxy base URL down to its `/v1` root,
+ * mirroring `sanitizeAnthropicBaseUrl`/`sanitizeClaudeCodeCompatibleBaseUrl`:
+ * strip trailing slash + known OpenAI-compatible suffixes, then ensure the
+ * result ends at `/v1` (append if absent, guard against double `/v1/v1`).
+ */
+function sanitizeVibeProxyBaseUrl(baseUrl: string) {
+  let base = (baseUrl || "")
+    .trim()
+    .replace(/\/$/, "")
+    .replace(/\/chat\/completions$/i, "")
+    .replace(/\/completions$/i, "");
+  // Guard against a literal "scheme://v1" authority so we never strip the host itself.
+  if (base.endsWith("/v1") && !base.endsWith("://v1")) {
+    return base;
+  }
+  return `${base}/v1`;
+}
+
 function sanitizeAnthropicBaseUrl(baseUrl: string) {
   return (baseUrl || "")
     .trim()
@@ -76,16 +104,40 @@ export async function POST(request) {
       baseUrl,
       type,
       compatMode,
+      preset,
       chatPath,
       modelsPath,
       customHeaders,
       iconUrl,
     } = validation.data;
 
+    if (preset === "vibeproxy-openai") {
+      // Schema guarantees baseUrl is non-empty for this preset.
+      const sanitizedBaseUrl = sanitizeVibeProxyBaseUrl(baseUrl as string);
+      const baseUrlError = validateProviderNodeBaseUrl(sanitizedBaseUrl);
+      if (baseUrlError) return baseUrlError;
+
+      const node = await createProviderNode({
+        id: `${OPENAI_COMPATIBLE_PREFIX}${VIBEPROXY_OPENAI_DEFAULTS.apiType}-${generateId()}`,
+        type: "openai-compatible",
+        prefix: (prefix?.trim() || VIBEPROXY_OPENAI_DEFAULTS.prefix).trim(),
+        apiType: apiType || VIBEPROXY_OPENAI_DEFAULTS.apiType,
+        baseUrl: sanitizedBaseUrl,
+        name: (name?.trim() || VIBEPROXY_OPENAI_DEFAULTS.name).trim(),
+        chatPath: chatPath || null,
+        modelsPath: modelsPath || null,
+        iconUrl: iconUrl?.trim() || null,
+        customHeaders: customHeaders || null,
+      });
+      return NextResponse.json({ node }, { status: 201 });
+    }
+
     // Determine type
     const nodeType = type || "openai-compatible";
 
     if (nodeType === "openai-compatible") {
+      const resolvedName = (name || "").trim();
+      const resolvedPrefix = (prefix || "").trim();
       const resolvedBaseUrl = (baseUrl || OPENAI_COMPATIBLE_DEFAULTS.baseUrl).trim();
       const baseUrlError = validateProviderNodeBaseUrl(resolvedBaseUrl);
       if (baseUrlError) return baseUrlError;
@@ -93,10 +145,10 @@ export async function POST(request) {
       const node = await createProviderNode({
         id: `${OPENAI_COMPATIBLE_PREFIX}${apiType}-${generateId()}`,
         type: "openai-compatible",
-        prefix: prefix.trim(),
+        prefix: resolvedPrefix,
         apiType,
         baseUrl: resolvedBaseUrl,
-        name: name.trim(),
+        name: resolvedName,
         chatPath: chatPath || null,
         modelsPath: modelsPath || null,
         iconUrl: iconUrl?.trim() || null,
@@ -124,9 +176,9 @@ export async function POST(request) {
             ? `${CLAUDE_CODE_COMPATIBLE_PREFIX}${generateId()}`
             : `${ANTHROPIC_COMPATIBLE_PREFIX}${generateId()}`,
         type: "anthropic-compatible",
-        prefix: prefix.trim(),
+        prefix: (prefix || "").trim(),
         baseUrl: sanitizedBaseUrl,
-        name: name.trim(),
+        name: (name || "").trim(),
         chatPath: chatPath || null,
         modelsPath: compatMode === "cc" ? null : modelsPath || null,
         iconUrl: iconUrl?.trim() || null,
diff --git a/src/shared/constants/providers/audio.ts b/src/shared/constants/providers/audio.ts
index 3b4376e61e..5a6d9cca1c 100644
--- a/src/shared/constants/providers/audio.ts
+++ b/src/shared/constants/providers/audio.ts
@@ -87,4 +87,16 @@ export const AUDIO_ONLY_PROVIDERS = {
     textIcon: "RV",
     website: "https://www.rev.ai",
   },
+
+  speechmatics: {
+    id: "speechmatics",
+    alias: "sm",
+    name: "Speechmatics",
+    icon: "record_voice_over",
+    color: "#0A2540",
+    textIcon: "SM",
+    website: "https://www.speechmatics.com",
+    hasFree: true,
+    freeNote: "Free tier — 8 hours/month, no credit card required. Batch (async) mode only.",
+  },
 };
diff --git a/src/shared/validation/schemas/provider.ts b/src/shared/validation/schemas/provider.ts
index 456d30902e..95ee820d55 100644
--- a/src/shared/validation/schemas/provider.ts
+++ b/src/shared/validation/schemas/provider.ts
@@ -270,8 +270,11 @@ export const removeModelAliasSchema = z.object({
 
 export const createProviderNodeSchema = z
   .object({
-    name: z.string().trim().min(1, "Name is required"),
-    prefix: z.string().trim().min(1, "Prefix is required"),
+    // #6874: name/prefix are required in general, but a `preset` (e.g.
+    // "vibeproxy-openai") supplies both — enforced conditionally below
+    // instead of unconditionally here.
+    name: z.string().trim().optional().or(z.literal("")),
+    prefix: z.string().trim().optional().or(z.literal("")),
     apiType: z
       .enum([
         "chat",
@@ -285,6 +288,10 @@ export const createProviderNodeSchema = z
     baseUrl: z.string().trim().min(1).optional(),
     type: z.enum(["openai-compatible", "anthropic-compatible"]).optional(),
     compatMode: z.enum(["cc"]).optional(),
+    // #6874: named presets fill in name/prefix/apiType for well-known
+    // OpenAI-compatible local gateways so the operator only has to paste
+    // a baseUrl. Currently just VibeProxy (github.com/automazeio/vibeproxy).
+    preset: z.enum(["vibeproxy-openai"]).optional(),
     chatPath: z.string().trim().startsWith("/").max(500).optional().or(z.literal("")),
     modelsPath: z.string().trim().startsWith("/").max(500).optional().or(z.literal("")),
     // #2166: optional operator-supplied remote icon URL for the provider node. Empty
@@ -296,6 +303,33 @@ export const createProviderNodeSchema = z
   })
   .superRefine((value, ctx) => {
     const nodeType = value.type || "openai-compatible";
+    if (value.preset === "vibeproxy-openai") {
+      // Preset supplies name/prefix/apiType — but baseUrl is still mandatory
+      // (a local proxy's host/port is operator-specific, unlike the generic
+      // openai-compatible fallback-to-api.openai.com default).
+      if (!value.baseUrl || !value.baseUrl.trim()) {
+        ctx.addIssue({
+          code: z.ZodIssueCode.custom,
+          message: "Base URL is required for the VibeProxy preset",
+          path: ["baseUrl"],
+        });
+      }
+      return;
+    }
+    if (!value.name || !value.name.trim()) {
+      ctx.addIssue({
+        code: z.ZodIssueCode.custom,
+        message: "Name is required",
+        path: ["name"],
+      });
+    }
+    if (!value.prefix || !value.prefix.trim()) {
+      ctx.addIssue({
+        code: z.ZodIssueCode.custom,
+        message: "Prefix is required",
+        path: ["prefix"],
+      });
+    }
     if (nodeType === "openai-compatible" && !value.apiType) {
       ctx.addIssue({
         code: z.ZodIssueCode.custom,
diff --git a/tests/unit/gtts-provider.test.ts b/tests/unit/gtts-provider.test.ts
new file mode 100644
index 0000000000..337d80ffa1
--- /dev/null
+++ b/tests/unit/gtts-provider.test.ts
@@ -0,0 +1,187 @@
+// gTTS (Google Translate TTS) audio-tts provider (#6667).
+//
+// The originally-proposed `/translate_tts` GET endpoint is deprecated;
+// this covers the current `batchexecute` RPC mechanism: chunking to the
+// 100-char limit, RPC body construction, response parsing, registry lookup,
+// and the handler's dispatch + error path.
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import {
+  GOOGLE_TTS_MAX_CHARS,
+  GttsUpstreamError,
+  buildGttsRpcBody,
+  chunkGttsText,
+  normalizeGttsLang,
+  parseBatchExecuteResponse,
+  synthesizeGtts,
+} from "../../open-sse/executors/gtts.ts";
+import { getSpeechProvider, parseSpeechModel } from "../../open-sse/config/audioRegistry.ts";
+import { handleAudioSpeech } from "../../open-sse/handlers/audioSpeech.ts";
+
+// ─── chunkGttsText ──────────────────────────────────────────────────────
+
+test("chunkGttsText returns a single chunk for short input", () => {
+  assert.deepEqual(chunkGttsText("hello world"), ["hello world"]);
+});
+
+test("chunkGttsText returns [] for empty/whitespace-only input", () => {
+  assert.deepEqual(chunkGttsText(""), []);
+  assert.deepEqual(chunkGttsText("   "), []);
+  assert.deepEqual(chunkGttsText(undefined), []);
+});
+
+test("chunkGttsText splits long input on whitespace within the 100-char limit", () => {
+  const text = Array.from({ length: 20 }, (_, i) => `word${i}`).join(" "); // > 100 chars
+  const chunks = chunkGttsText(text);
+  assert.ok(chunks.length > 1, "must split into multiple chunks");
+  for (const chunk of chunks) {
+    assert.ok(chunk.length <= GOOGLE_TTS_MAX_CHARS, `chunk exceeds limit: "${chunk}"`);
+  }
+  // Rejoining chunks must reconstruct the original words (no character loss).
+  assert.equal(chunks.join(" "), text);
+});
+
+test("chunkGttsText hard-splits a single word longer than the limit", () => {
+  const longWord = "a".repeat(250);
+  const chunks = chunkGttsText(longWord, 100);
+  assert.equal(chunks.length, 3);
+  assert.equal(chunks.join(""), longWord);
+  for (const chunk of chunks) {
+    assert.ok(chunk.length <= 100);
+  }
+});
+
+// ─── normalizeGttsLang ──────────────────────────────────────────────────
+
+test("normalizeGttsLang accepts simple language codes and defaults to English", () => {
+  assert.equal(normalizeGttsLang("pt-BR"), "pt-BR");
+  assert.equal(normalizeGttsLang("es"), "es");
+  assert.equal(normalizeGttsLang(""), "en");
+  assert.equal(normalizeGttsLang(undefined), "en");
+  assert.equal(normalizeGttsLang("<script>alert(1)</script>"), "en");
+});
+
+// ─── buildGttsRpcBody ───────────────────────────────────────────────────
+
+test("buildGttsRpcBody wraps text/lang under the jQ1olc RPC id, urlencoded", () => {
+  const body = buildGttsRpcBody("hello", "en");
+  assert.match(body, /^f\.req=/);
+  assert.ok(body.endsWith("&"));
+
+  const encoded = body.slice("f.req=".length, -1);
+  const envelope = JSON.parse(decodeURIComponent(encoded));
+  assert.deepEqual(envelope, [[["jQ1olc", JSON.stringify(["hello", "en", true, "null"]), null, "generic"]]]);
+});
+
+// ─── parseBatchExecuteResponse ──────────────────────────────────────────
+
+function buildBatchExecuteFixture(base64Audio: string): string {
+  const inner = JSON.stringify([base64Audio, null, null, null, null, null, []]);
+  const outer = JSON.stringify([["wrb.fr", "jQ1olc", inner, null, null, null, "generic"]]);
+  return `)]}'\n\n${outer.length}\n${outer}\n`;
+}
+
+test("parseBatchExecuteResponse extracts the base64 audio payload", () => {
+  const raw = buildBatchExecuteFixture("aGVsbG8=");
+  assert.equal(parseBatchExecuteResponse(raw), "aGVsbG8=");
+});
+
+test("parseBatchExecuteResponse throws GttsUpstreamError when no payload is found", () => {
+  assert.throws(() => parseBatchExecuteResponse(")]}'\n\nnot json at all"), GttsUpstreamError);
+});
+
+// ─── synthesizeGtts (network via injectable fetch) ───────────────────────
+
+test("synthesizeGtts concatenates decoded audio across multiple chunks", async () => {
+  const text = Array.from({ length: 20 }, (_, i) => `word${i}`).join(" ");
+  const expectedChunks = chunkGttsText(text);
+  assert.ok(expectedChunks.length > 1);
+
+  const calls: string[] = [];
+  const fetchImpl = async (url: string, init: RequestInit) => {
+    calls.push(String(init.body));
+    return new Response(buildBatchExecuteFixture(Buffer.from("chunk-audio").toString("base64")), {
+      status: 200,
+      headers: { "content-type": "text/plain" },
+    });
+  };
+
+  const audio = await synthesizeGtts({ text, lang: "en" }, fetchImpl);
+  assert.equal(calls.length, expectedChunks.length);
+  assert.equal(audio.toString(), "chunk-audio".repeat(expectedChunks.length));
+});
+
+test("synthesizeGtts throws GttsUpstreamError with the upstream status on a non-ok response", async () => {
+  const fetchImpl = async () =>
+    new Response("rate limited", { status: 429 });
+
+  await assert.rejects(
+    () => synthesizeGtts({ text: "hi", lang: "en" }, fetchImpl),
+    (err: unknown) => err instanceof GttsUpstreamError && err.status === 429
+  );
+});
+
+// ─── registry lookup ──────────────────────────────────────────────────
+
+test("gtts is registered as a no-auth speech provider", () => {
+  const provider = getSpeechProvider("gtts");
+  assert.ok(provider);
+  assert.equal(provider?.authType, "none");
+  assert.equal(provider?.format, "gtts");
+
+  const parsed = parseSpeechModel("gtts/default");
+  assert.equal(parsed.provider, "gtts");
+  assert.equal(parsed.model, "default");
+});
+
+// ─── handleAudioSpeech dispatch ─────────────────────────────────────────
+
+test("handleAudioSpeech routes gtts requests without requiring credentials", async () => {
+  const originalFetch = globalThis.fetch;
+  globalThis.fetch = (async (_url: string, init: RequestInit) => {
+    assert.match(String(init.body), /^f\.req=/);
+    return new Response(buildBatchExecuteFixture(Buffer.from("hi-audio").toString("base64")), {
+      status: 200,
+      headers: { "content-type": "text/plain" },
+    });
+  }) as typeof fetch;
+
+  try {
+    const providerConfig = getSpeechProvider("gtts");
+    const response = await handleAudioSpeech({
+      body: { model: "gtts/default", input: "hi there", voice: "en" },
+      credentials: null,
+      resolvedProvider: providerConfig,
+      resolvedModel: "default",
+    });
+
+    assert.equal(response.status, 200);
+    assert.equal(response.headers.get("content-type"), "audio/mpeg");
+    const buf = Buffer.from(await response.arrayBuffer());
+    assert.equal(buf.toString(), "hi-audio");
+  } finally {
+    globalThis.fetch = originalFetch;
+  }
+});
+
+test("handleAudioSpeech surfaces gtts upstream errors without leaking a stack trace", async () => {
+  const originalFetch = globalThis.fetch;
+  globalThis.fetch = (async () => new Response("blocked", { status: 429 })) as typeof fetch;
+
+  try {
+    const providerConfig = getSpeechProvider("gtts");
+    const response = await handleAudioSpeech({
+      body: { model: "gtts/default", input: "hi there" },
+      credentials: null,
+      resolvedProvider: providerConfig,
+      resolvedModel: "default",
+    });
+    const payload = (await response.json()) as { error: { message: string } };
+
+    assert.equal(response.status, 429);
+    assert.ok(!payload.error.message.includes("at /"), "error body must not leak a stack trace");
+  } finally {
+    globalThis.fetch = originalFetch;
+  }
+});
diff --git a/tests/unit/provider-nodes-vibeproxy-preset.test.ts b/tests/unit/provider-nodes-vibeproxy-preset.test.ts
new file mode 100644
index 0000000000..59830e8650
--- /dev/null
+++ b/tests/unit/provider-nodes-vibeproxy-preset.test.ts
@@ -0,0 +1,135 @@
+// #6874 — VibeProxy (github.com/automazeio/vibeproxy) provider-node preset.
+// TDD regression guard for `preset: "vibeproxy-openai"` on
+// POST /api/provider-nodes: defaults name/prefix/apiType, still requires an
+// explicit baseUrl, and normalizes the caller-supplied baseUrl to its `/v1`
+// root the same way the existing Anthropic/Claude-Code-compatible presets do.
+import test from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-vibeproxy-preset-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const core = await import("../../src/lib/db/core.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const providerNodesRoute = await import("../../src/app/api/provider-nodes/route.ts");
+const { OPENAI_COMPATIBLE_PREFIX } = await import("../../src/shared/constants/providers.ts");
+
+interface ProviderNodeErrorBody {
+  error: { message: string; details?: { field: string; message: string }[] };
+}
+
+interface ProviderNodeResponseBody {
+  node: {
+    id: string;
+    type: string;
+    prefix: string;
+    name: string;
+    apiType?: string;
+    baseUrl: string;
+  };
+}
+
+async function resetStorage() {
+  delete process.env.OMNIROUTE_ALLOW_PRIVATE_PROVIDER_URLS;
+  delete process.env.OMNIROUTE_ALLOW_LOCAL_PROVIDER_URLS;
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+  fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
+}
+
+function makeRequest(body: unknown) {
+  return new Request("http://localhost/api/provider-nodes", {
+    method: "POST",
+    headers: { "content-type": "application/json" },
+    body: JSON.stringify(body),
+  });
+}
+
+test.beforeEach(async () => {
+  await resetStorage();
+});
+
+test.after(async () => {
+  await resetStorage();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+test("vibeproxy-openai preset creates a node with defaulted name/prefix/apiType", async () => {
+  const response = await providerNodesRoute.POST(
+    makeRequest({
+      preset: "vibeproxy-openai",
+      baseUrl: "http://localhost:8317",
+    })
+  );
+  const body = (await response.json()) as ProviderNodeResponseBody;
+
+  assert.equal(response.status, 201);
+  assert.match(body.node.id, new RegExp(`^${OPENAI_COMPATIBLE_PREFIX}chat-`));
+  assert.equal(body.node.type, "openai-compatible");
+  assert.equal(body.node.name, "VibeProxy");
+  assert.equal(body.node.prefix, "vibeproxy");
+  assert.equal(body.node.apiType, "chat");
+  assert.equal(body.node.baseUrl, "http://localhost:8317/v1");
+});
+
+test("vibeproxy-openai preset normalizes a baseUrl with a /chat/completions suffix", async () => {
+  const response = await providerNodesRoute.POST(
+    makeRequest({
+      preset: "vibeproxy-openai",
+      baseUrl: "http://localhost:8317/v1/chat/completions",
+    })
+  );
+  const body = (await response.json()) as ProviderNodeResponseBody;
+
+  assert.equal(response.status, 201);
+  assert.equal(body.node.baseUrl, "http://localhost:8317/v1");
+});
+
+test("vibeproxy-openai preset appends /v1 when the caller omits it", async () => {
+  const response = await providerNodesRoute.POST(
+    makeRequest({
+      preset: "vibeproxy-openai",
+      baseUrl: "http://localhost:9000",
+    })
+  );
+  const body = (await response.json()) as ProviderNodeResponseBody;
+
+  assert.equal(response.status, 201);
+  assert.equal(body.node.baseUrl, "http://localhost:9000/v1");
+});
+
+test("vibeproxy-openai preset honors caller-supplied name/prefix instead of the defaults", async () => {
+  const response = await providerNodesRoute.POST(
+    makeRequest({
+      preset: "vibeproxy-openai",
+      name: "My VibeProxy",
+      prefix: "my-vibeproxy",
+      baseUrl: "http://localhost:8317",
+    })
+  );
+  const body = (await response.json()) as ProviderNodeResponseBody;
+
+  assert.equal(response.status, 201);
+  assert.equal(body.node.name, "My VibeProxy");
+  assert.equal(body.node.prefix, "my-vibeproxy");
+});
+
+test("vibeproxy-openai preset rejects a missing baseUrl (no silent default)", async () => {
+  const response = await providerNodesRoute.POST(
+    makeRequest({
+      preset: "vibeproxy-openai",
+    })
+  );
+  const body = (await response.json()) as ProviderNodeErrorBody;
+
+  assert.equal(response.status, 400);
+  assert.equal(body.error.message, "Invalid request");
+  assert.match(
+    body.error.details?.find((d) => d.field === "baseUrl")?.message || "",
+    /Base URL is required for the VibeProxy preset/
+  );
+  assert.deepEqual(await providersDb.getProviderNodes(), []);
+});
diff --git a/tests/unit/speechmatics-transcription-provider.test.ts b/tests/unit/speechmatics-transcription-provider.test.ts
new file mode 100644
index 0000000000..cf61fb09ea
--- /dev/null
+++ b/tests/unit/speechmatics-transcription-provider.test.ts
@@ -0,0 +1,143 @@
+// #6659 — Speechmatics STT provider: async batch workflow (submit multipart
+// job → poll → fetch transcript), mirroring the existing AssemblyAI/Rev AI
+// adapters. Streaming (WebSocket) mode is explicitly out of scope for v1.
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { handleAudioTranscription } = await import("../../open-sse/handlers/audioTranscription.ts");
+const { getTranscriptionProvider } = await import("../../open-sse/config/audioRegistry.ts");
+
+function buildFile(contents: string, name: string, type: string) {
+  return new File([Buffer.from(contents)], name, { type });
+}
+
+function immediateTimeout(callback: (...args: unknown[]) => void, _ms?: number, ...args: unknown[]) {
+  if (typeof callback === "function") callback(...args);
+  return 0;
+}
+
+test("speechmatics registry entry is async, apikey-bearer, batch-only", () => {
+  const provider = getTranscriptionProvider("speechmatics");
+  assert.ok(provider);
+  assert.equal(provider?.authType, "apikey");
+  assert.equal(provider?.authHeader, "bearer");
+  assert.equal(provider?.async, true);
+  assert.equal(provider?.format, "speechmatics");
+  assert.ok(provider?.models.some((m) => m.id === "enhanced"));
+});
+
+test("handleAudioTranscription routes Speechmatics: submit job → poll → fetch transcript", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalSetTimeout = globalThis.setTimeout;
+  const calls: { url: string; method: string }[] = [];
+
+  // @ts-expect-error test double swaps the timer signature intentionally
+  globalThis.setTimeout = immediateTimeout;
+  globalThis.fetch = (async (url: string, options: RequestInit = {}) => {
+    const stringUrl = String(url);
+    calls.push({ url: stringUrl, method: (options?.method as string) || "GET" });
+
+    if (stringUrl === "https://asr.api.speechmatics.com/v2/jobs") {
+      assert.equal(options.method, "POST");
+      assert.equal((options.headers as Record<string, string>).Authorization, "Bearer sm-key");
+      return new Response(JSON.stringify({ id: "job-1" }), {
+        status: 200,
+        headers: { "content-type": "application/json" },
+      });
+    }
+
+    if (stringUrl === "https://asr.api.speechmatics.com/v2/jobs/job-1") {
+      return new Response(JSON.stringify({ job: { status: "done" } }), {
+        status: 200,
+        headers: { "content-type": "application/json" },
+      });
+    }
+
+    if (stringUrl === "https://asr.api.speechmatics.com/v2/jobs/job-1/transcript?format=txt") {
+      return new Response("speechmatics result", {
+        status: 200,
+        headers: { "content-type": "text/plain" },
+      });
+    }
+
+    throw new Error(`Unexpected URL: ${stringUrl}`);
+  }) as typeof fetch;
+
+  try {
+    const formData = new FormData();
+    formData.append("model", "speechmatics/enhanced");
+    formData.append("file", buildFile("abc", "clip.wav", "audio/wav"));
+
+    const response = await handleAudioTranscription({
+      formData,
+      credentials: { apiKey: "sm-key" },
+    });
+
+    assert.deepEqual(await response.json(), { text: "speechmatics result" });
+    assert.deepEqual(
+      calls.map((entry) => entry.url),
+      [
+        "https://asr.api.speechmatics.com/v2/jobs",
+        "https://asr.api.speechmatics.com/v2/jobs/job-1",
+        "https://asr.api.speechmatics.com/v2/jobs/job-1/transcript?format=txt",
+      ]
+    );
+  } finally {
+    globalThis.fetch = originalFetch;
+    globalThis.setTimeout = originalSetTimeout;
+  }
+});
+
+test("handleAudioTranscription returns an error when Speechmatics rejects the job", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalSetTimeout = globalThis.setTimeout;
+
+  // @ts-expect-error test double swaps the timer signature intentionally
+  globalThis.setTimeout = immediateTimeout;
+  globalThis.fetch = (async (url: string, options: RequestInit = {}) => {
+    const stringUrl = String(url);
+    if (stringUrl === "https://asr.api.speechmatics.com/v2/jobs") {
+      return new Response(JSON.stringify({ id: "job-2" }), {
+        status: 200,
+        headers: { "content-type": "application/json" },
+      });
+    }
+    if (stringUrl === "https://asr.api.speechmatics.com/v2/jobs/job-2") {
+      return new Response(
+        JSON.stringify({ job: { status: "rejected", errors: [{ message: "bad audio" }] } }),
+        { status: 200, headers: { "content-type": "application/json" } }
+      );
+    }
+    throw new Error(`Unexpected URL: ${stringUrl} (${options.method})`);
+  }) as typeof fetch;
+
+  try {
+    const formData = new FormData();
+    formData.append("model", "speechmatics/enhanced");
+    formData.append("file", buildFile("abc", "clip.wav", "audio/wav"));
+
+    const response = await handleAudioTranscription({
+      formData,
+      credentials: { apiKey: "sm-key" },
+    });
+    const payload = (await response.json()) as { error: { message: string } };
+
+    assert.equal(response.status, 500);
+    assert.equal(payload.error.message, "bad audio");
+  } finally {
+    globalThis.fetch = originalFetch;
+    globalThis.setTimeout = originalSetTimeout;
+  }
+});
+
+test("handleAudioTranscription requires credentials for Speechmatics", async () => {
+  const formData = new FormData();
+  formData.append("model", "speechmatics/enhanced");
+  formData.append("file", buildFile("abc", "clip.wav", "audio/wav"));
+
+  const response = await handleAudioTranscription({ formData, credentials: null });
+  const payload = (await response.json()) as { error: { message: string } };
+
+  assert.equal(response.status, 401);
+  assert.match(payload.error.message, /speechmatics/);
+});