@todoforai/voiceloop 0.1.22 → 0.1.23
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/examples/token-worker/.wrangler/cache/wrangler-account.json +6 -0
- package/examples/token-worker/README.md +2 -0
- package/examples/token-worker/index.js +15 -2
- package/examples/token-worker/wrangler.toml +1 -1
- package/index.html +14 -2
- package/package.json +1 -1
- package/src/voice-agent.js +10 -5
|
@@ -7,6 +7,7 @@ out only what each hop needs.
|
|
|
7
7
|
| route | returns | why |
|
|
8
8
|
|---|---|---|
|
|
9
9
|
| `POST /stt/token` | `{ token, expires_in }` | short-TTL Deepgram key; the browser opens the WS itself |
|
|
10
|
+
| `POST /stt/soniox-token` | `{ token, expires_in }` | short-TTL Soniox key (same shape) |
|
|
10
11
|
| `POST /tts` | `audio/mpeg` | ElevenLabs proxy — the `xi-api-key` stays server-side |
|
|
11
12
|
| `POST /llm` | OpenAI-format SSE | Anthropic proxy — Claude has no short-TTL key to mint, so the key stays here |
|
|
12
13
|
|
|
@@ -17,6 +18,7 @@ out only what each hop needs.
|
|
|
17
18
|
```sh
|
|
18
19
|
wrangler deploy
|
|
19
20
|
wrangler secret put DEEPGRAM_API_KEY
|
|
21
|
+
wrangler secret put SONIOX_API_KEY
|
|
20
22
|
wrangler secret put ELEVENLABS_API_KEY
|
|
21
23
|
wrangler secret put ANTHROPIC_API_KEY
|
|
22
24
|
```
|
|
@@ -4,10 +4,11 @@
|
|
|
4
4
|
// never reach the page, so this Worker holds them and hands out only what each hop needs:
|
|
5
5
|
//
|
|
6
6
|
// POST /stt/token → { token, expires_in } short-TTL Deepgram key (browser opens the WS itself)
|
|
7
|
+
// POST /stt/soniox-token → same shape short-TTL Soniox key
|
|
7
8
|
// POST /tts → audio/mpeg bytes ElevenLabs proxy (the xi-api-key stays here)
|
|
8
9
|
// POST /llm → OpenAI-format SSE Anthropic proxy (there is no short-TTL key to mint)
|
|
9
10
|
//
|
|
10
|
-
// Secrets (wrangler secret put): DEEPGRAM_API_KEY, ELEVENLABS_API_KEY, ANTHROPIC_API_KEY.
|
|
11
|
+
// Secrets (wrangler secret put): DEEPGRAM_API_KEY, SONIOX_API_KEY, ELEVENLABS_API_KEY, ANTHROPIC_API_KEY.
|
|
11
12
|
// ALLOWED_ORIGINS (var) is a comma-separated allowlist — these routes spend real money, so an
|
|
12
13
|
// open CORS policy is an open wallet.
|
|
13
14
|
//
|
|
@@ -99,7 +100,7 @@ export default {
|
|
|
99
100
|
// Everything below mints a credential or bills a provider. /stt/token mints at most one
|
|
100
101
|
// credential per 5 min so it takes the tight limiter; /tts fires once per spoken SENTENCE and
|
|
101
102
|
// /llm once per turn plus prefetch, so the conversation hot path gets its own, roomier window.
|
|
102
|
-
const limiter = url.pathname
|
|
103
|
+
const limiter = url.pathname.startsWith('/stt/') ? env.PAID_RATE_LIMITER : env.CONVO_RATE_LIMITER;
|
|
103
104
|
if (await rateLimited(req, env, limiter)) return json({ error: 'rate limited — try again in a minute' }, 429, origin);
|
|
104
105
|
|
|
105
106
|
if (url.pathname === '/stt/token') {
|
|
@@ -115,6 +116,18 @@ export default {
|
|
|
115
116
|
return json({ token: body.access_token, expires_in: body.expires_in ?? 300 }, 200, origin);
|
|
116
117
|
}
|
|
117
118
|
|
|
119
|
+
if (url.pathname === '/stt/soniox-token') {
|
|
120
|
+
if (!env.SONIOX_API_KEY) return json({ error: 'SONIOX_API_KEY not configured' }, 500, origin);
|
|
121
|
+
const r = await fetch('https://api.soniox.com/v1/auth/temporary-api-key', {
|
|
122
|
+
method: 'POST',
|
|
123
|
+
headers: { authorization: `Bearer ${env.SONIOX_API_KEY}`, 'content-type': 'application/json' },
|
|
124
|
+
body: JSON.stringify({ usage_type: 'transcribe_websocket', expires_in_seconds: 300 }),
|
|
125
|
+
});
|
|
126
|
+
const body = await r.json().catch(() => ({}));
|
|
127
|
+
if (!r.ok || !body.api_key) return json({ error: body.message || body.error || 'mint failed' }, 502, origin);
|
|
128
|
+
return json({ token: body.api_key, expires_in: 300 }, 200, origin);
|
|
129
|
+
}
|
|
130
|
+
|
|
118
131
|
if (url.pathname === '/tts') {
|
|
119
132
|
if (!env.ELEVENLABS_API_KEY) return json({ error: 'ELEVENLABS_API_KEY not configured' }, 500, origin);
|
|
120
133
|
const { text, voice_id, model_id, output_format } = await req.json().catch(() => ({}));
|
package/index.html
CHANGED
|
@@ -318,13 +318,24 @@
|
|
|
318
318
|
</main>
|
|
319
319
|
|
|
320
320
|
<script type="module">
|
|
321
|
-
import { VoiceAgent, PiperTTS, ElevenLabsTTS, makeOpenAILLM, unlockAudio, prewarmVoice, STT_PROVIDERS, sttSelfCaptures, LANG_NAMES } from './src/index.js';
|
|
321
|
+
import { VoiceAgent, PiperTTS, ElevenLabsTTS, makeOpenAILLM, unlockAudio, prewarmVoice, warmDeepgramToken, warmSonioxToken, STT_PROVIDERS, sttSelfCaptures, LANG_NAMES } from './src/index.js';
|
|
322
322
|
|
|
323
323
|
// Load VAD/WASM + build the Silero session while the page sits idle — but ONLY for a provider that
|
|
324
324
|
// will actually use it. A selfCapture provider (webspeech) builds no pipeline, so prewarming there
|
|
325
325
|
// downloads an ONNX model to feed a detector that never runs.
|
|
326
326
|
const prewarmIfPipelined = () => { if (!sttSelfCaptures($('sttProvider').value)) prewarmVoice(); };
|
|
327
327
|
|
|
328
|
+
// Pay the per-hop cold costs while idle, not on ▶: the STT token mint (0.6–2.7s when it races the
|
|
329
|
+
// mic/VAD start) and the LLM route's DNS+TLS. Both keyed on the current field values so a change
|
|
330
|
+
// re-warms the NEW target; all best-effort — open() re-mints and the first turn reconnects on miss.
|
|
331
|
+
const STT_TOKEN_WARMERS = { deepgram: warmDeepgramToken, soniox: warmSonioxToken };
|
|
332
|
+
const prewarmRoutes = () => {
|
|
333
|
+
const warm = STT_TOKEN_WARMERS[$('sttProvider').value], tokenUrl = $('sttTokenUrl').value.trim();
|
|
334
|
+
if (warm && tokenUrl) warm(tokenUrl, $('apiKey').value.trim()).catch(() => {});
|
|
335
|
+
const llmUrl = $('mode').value === 'llm' ? $('llmUrl').value.trim() : $('mode').value === 'hosted' ? HOSTED_LLM_URL : '';
|
|
336
|
+
if (llmUrl) fetch(llmUrl, { method: 'HEAD', signal: AbortSignal.timeout(3000) }).catch(() => {});
|
|
337
|
+
};
|
|
338
|
+
|
|
328
339
|
const $ = id => document.getElementById(id);
|
|
329
340
|
const fill = (id, entries) => $(id).append(...entries.map(([value, textContent]) =>
|
|
330
341
|
Object.assign(document.createElement('option'), { value, textContent })));
|
|
@@ -461,7 +472,7 @@ function onSettingChanged() {
|
|
|
461
472
|
for (const id of ['llmUrl', 'model']) $(id).disabled = !custom;
|
|
462
473
|
$('sttTokenUrl').disabled = self;
|
|
463
474
|
$('apiKey').disabled = !custom && self; // used as LLM Bearer OR token-route X-API-Key — dead only when neither hop exists
|
|
464
|
-
showPipeline(); prewarmIfPipelined();
|
|
475
|
+
showPipeline(); prewarmIfPipelined(); prewarmRoutes();
|
|
465
476
|
}
|
|
466
477
|
|
|
467
478
|
// ── the orb's level ─────────────────────────────────────────────────────────
|
|
@@ -743,6 +754,7 @@ const applyTts = () => {
|
|
|
743
754
|
wireModelProgress();
|
|
744
755
|
dropAgent('tts');
|
|
745
756
|
loadVoices();
|
|
757
|
+
if (tts instanceof ElevenLabsTTS) tts.warm().catch(() => {}); // DNS+TLS now, not 2.5s after ▶ (Piper's 60MB stays behind the click)
|
|
746
758
|
log('tts →', $('ttsProvider').value);
|
|
747
759
|
};
|
|
748
760
|
let ttsUrlTimer;
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@todoforai/voiceloop",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.23",
|
|
4
4
|
"description": "Fast, interruptible browser voice agent: VAD → STT → LLM → TTS with real barge-in, echo filtering, streaming first-sentence TTS and tap-to-seek. Zero runtime dependencies.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "src/index.js",
|
package/src/voice-agent.js
CHANGED
|
@@ -407,11 +407,16 @@ export class VoiceAgent {
|
|
|
407
407
|
try {
|
|
408
408
|
stream = await navigator.mediaDevices.getUserMedia({ audio });
|
|
409
409
|
} catch (e) {
|
|
410
|
-
//
|
|
411
|
-
//
|
|
412
|
-
//
|
|
413
|
-
|
|
414
|
-
|
|
410
|
+
// OverconstrainedError ("Invalid constraint" on WebKit) has two causes:
|
|
411
|
+
// - the saved mic id no longer exists here (unplugged / different device / stale
|
|
412
|
+
// cross-profile localStorage) — `exact` rejects instead of falling back;
|
|
413
|
+
// - WebKitGTK (Tauri on Linux) treats the bare DSP constraints as REQUIRED, so
|
|
414
|
+
// channelCount/autoGainControl can be unsatisfiable even with no deviceId.
|
|
415
|
+
// Degrade in that order — a default mic with default DSP beats no session at all.
|
|
416
|
+
if (e.name !== 'OverconstrainedError') throw e;
|
|
417
|
+
stream = await navigator.mediaDevices.getUserMedia({ audio: { ...audio, deviceId: undefined } })
|
|
418
|
+
.catch(() => navigator.mediaDevices.getUserMedia({ audio: true }))
|
|
419
|
+
.catch(() => { throw e; });
|
|
415
420
|
}
|
|
416
421
|
} catch (e) {
|
|
417
422
|
// Mic open failed (permission denied / no device / constraints). We already optimistically
|