mikael-lovqvists-claude-agent
db8889aeed
STT (Silero VAD + Whisper via sherpa-onnx), Chatterbox TTS HTTP server, query completeness classifier (Ollama), multi-voice demo scripts, and planning docs. Kept as reference; clean rewrite planned in separate repos. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
86 lines
2.6 KiB
JavaScript
86 lines
2.6 KiB
JavaScript
/**
|
|
* Voice demo: microphone → Whisper STT → (optional LLM cleanup) → Bark TTS
|
|
*
|
|
* Bark supports emphasis via UPPERCASE, paralinguistic tokens ([laughs], [sighs],
|
|
* [clears throat]), and hesitation (...). Markdown is preprocessed automatically:
|
|
* **bold** → UPPERCASE, headers get a pause, etc.
|
|
*
|
|
* Environment variables:
|
|
* WHISPER_MODEL default: base.en
|
|
* options: tiny.en base.en small.en medium.en large-v3 large-v3-turbo
|
|
* BARK_MODEL default: suno/bark (use suno/bark-small for faster/smaller)
|
|
* BARK_VOICE default: v2/en_speaker_6
|
|
* options: v2/en_speaker_0 .. v2/en_speaker_9
|
|
* STT_PROVIDER default: cuda
|
|
* USE_LLM set to 0 to disable (default: 1)
|
|
* OLLAMA_MODEL default: phi3:mini
|
|
*/
|
|
|
|
import { Stt } from './lib/stt.mjs'
|
|
import { Bark_Tts } from './lib/bark-tts.mjs'
|
|
import { llm_available, list_models, cleanup } from './lib/llm.mjs'
|
|
|
|
const WHISPER_MODEL = process.env.WHISPER_MODEL || 'base.en'
|
|
const STT_PROVIDER = process.env.STT_PROVIDER || 'cuda'
|
|
const USE_LLM = process.env.USE_LLM !== '0'
|
|
|
|
process.stderr.write(`[stt] loading whisper-${WHISPER_MODEL} on ${STT_PROVIDER}...\n`)
|
|
const stt = new Stt({ whisper_name: WHISPER_MODEL, provider: STT_PROVIDER })
|
|
stt.init()
|
|
process.stderr.write('[stt] ready\n')
|
|
|
|
process.stderr.write('[tts] loading Bark (this takes a moment)...\n')
|
|
const tts = new Bark_Tts()
|
|
await tts.init()
|
|
process.stderr.write('[tts] Bark ready\n')
|
|
|
|
let has_llm = false
|
|
if (USE_LLM) {
|
|
has_llm = await llm_available()
|
|
if (has_llm) {
|
|
const models = await list_models()
|
|
process.stderr.write(`[llm] ollama available. models: ${models.slice(0, 6).join(', ')}\n`)
|
|
} else {
|
|
process.stderr.write('[llm] ollama not reachable, skipping cleanup\n')
|
|
}
|
|
}
|
|
|
|
process.stderr.write(`\n[ready] whisper=${WHISPER_MODEL} bark=${process.env.BARK_MODEL || 'suno/bark'} llm=${has_llm}\n`)
|
|
process.stderr.write('[ready] speak into your microphone. Ctrl+C to stop.\n\n')
|
|
|
|
let speaking = false
|
|
|
|
const stop = stt.listen(async (raw_text) => {
|
|
if (speaking) {
|
|
process.stderr.write(`[skipped] ${raw_text}\n`)
|
|
return
|
|
}
|
|
speaking = true
|
|
|
|
try {
|
|
process.stdout.write(`[raw] ${raw_text}\n`)
|
|
|
|
let text = raw_text
|
|
if (has_llm) {
|
|
text = await cleanup(raw_text)
|
|
if (text !== raw_text) {
|
|
process.stdout.write(`[llm] ${text}\n`)
|
|
}
|
|
}
|
|
|
|
await tts.speak_streaming(text)
|
|
process.stdout.write('\n')
|
|
} catch (err) {
|
|
process.stderr.write(`[error] ${err.message}\n`)
|
|
} finally {
|
|
speaking = false
|
|
}
|
|
})
|
|
|
|
process.on('SIGINT', () => {
|
|
process.stderr.write('\n[voice] stopping\n')
|
|
stop()
|
|
tts.stop()
|
|
process.exit(0)
|
|
})
|