feat: chatterbox TTS via madcat-tts daemon, Web Speech API STT, styled persona picker
- tts.py: replace piper subprocess with HTTP POST to madcat-tts /v1/audio/speech (chatterbox voice cloning) - chat.js: replace whisper server upload with browser Web Speech API (webkitSpeechRecognition) - chat.css: style persona picker — appearance:none select, themed with CSS vars, mobile responsive - main.py: default TTS voice → bt7274-en
This commit is contained in:
marauder-actual
+1
-1
@@ -403,7 +403,7 @@ else:
|
||||
|
||||
# --- TTS / STT ---
|
||||
TTS_ENABLED = os.environ.get("TTS_ENABLED", "true").lower() != "false"
|
||||
TTS_VOICE = os.environ.get("TTS_VOICE", "en_US-amy-medium")
|
||||
TTS_VOICE = os.environ.get("TTS_VOICE", "bt7274-en")
|
||||
tts = TTS(voice=TTS_VOICE) if TTS_ENABLED else None
|
||||
|
||||
STT_ENABLED = os.environ.get("STT_ENABLED", "true").lower() != "false"
|
||||
|
||||
@@ -242,6 +242,76 @@ body {
|
||||
.sigil:hover { opacity: 0.85; }
|
||||
.sigil img { width: 100%; height: 100%; display: block; }
|
||||
|
||||
/* ---------- persona picker (inline in topnav) ---------- */
|
||||
.topnav__persona-wrap {
|
||||
display: flex;
|
||||
align-items: center;
|
||||
gap: 0.45rem;
|
||||
}
|
||||
|
||||
.topnav__persona-label {
|
||||
font-family: var(--sans);
|
||||
font-size: 0.58rem;
|
||||
font-weight: 400;
|
||||
letter-spacing: 0.16em;
|
||||
text-transform: uppercase;
|
||||
color: var(--ink-faint);
|
||||
opacity: 0.45;
|
||||
cursor: default;
|
||||
user-select: none;
|
||||
}
|
||||
|
||||
.topnav__persona-select {
|
||||
appearance: none;
|
||||
-webkit-appearance: none;
|
||||
border: 1px solid var(--ink-faint);
|
||||
border-radius: 3px;
|
||||
background: var(--bg-soft);
|
||||
color: var(--ink-muted);
|
||||
font-family: var(--sans);
|
||||
font-size: 0.62rem;
|
||||
font-weight: 400;
|
||||
letter-spacing: 0.08em;
|
||||
text-transform: uppercase;
|
||||
padding: 0.2rem 1.3rem 0.2rem 0.45rem;
|
||||
cursor: pointer;
|
||||
outline: none;
|
||||
opacity: 0.55;
|
||||
transition: opacity 400ms ease, color 400ms ease, border-color 400ms ease;
|
||||
|
||||
/* custom dropdown arrow */
|
||||
background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='8' height='5' fill='none'%3E%3Cpath d='M1 1l3 3 3-3' stroke='%236a6a6f' stroke-width='1.2' stroke-linecap='round' stroke-linejoin='round'/%3E%3C/svg%3E");
|
||||
background-repeat: no-repeat;
|
||||
background-position: right 0.4rem center;
|
||||
}
|
||||
.topnav__persona-select:hover {
|
||||
opacity: 1;
|
||||
color: var(--ink);
|
||||
border-color: var(--ink-muted);
|
||||
}
|
||||
.topnav__persona-select:focus-visible {
|
||||
opacity: 1;
|
||||
color: var(--ink);
|
||||
border-color: var(--coral);
|
||||
}
|
||||
|
||||
.topnav__persona-status {
|
||||
font-family: var(--sans);
|
||||
font-size: 0.58rem;
|
||||
font-weight: 400;
|
||||
letter-spacing: 0.12em;
|
||||
color: var(--coral);
|
||||
opacity: 0.7;
|
||||
transition: opacity 400ms ease;
|
||||
user-select: none;
|
||||
}
|
||||
.topnav__persona-status:empty { display: none; }
|
||||
|
||||
@media (max-width: 600px) {
|
||||
.topnav__persona-label { display: none; }
|
||||
.topnav__persona-select { font-size: 0.58rem; padding: 0.15rem 1.1rem 0.15rem 0.35rem; }
|
||||
}
|
||||
|
||||
/* ===================== CONVERSATION ===================== */
|
||||
.conversation {
|
||||
flex: 1 0 auto;
|
||||
|
||||
+65
-55
@@ -241,77 +241,87 @@ document.addEventListener('keydown', (e) => {
|
||||
}
|
||||
});
|
||||
|
||||
// ---------- voice input (whisper) ----------
|
||||
// ---------- voice input (Web Speech API) ----------
|
||||
|
||||
const $mic = document.getElementById('mic-button');
|
||||
let mediaRecorder = null;
|
||||
let recordedChunks = [];
|
||||
const SpeechRecognition = window.SpeechRecognition || window.webkitSpeechRecognition;
|
||||
let recognition = null;
|
||||
let recording = false;
|
||||
|
||||
async function startRecording() {
|
||||
// Disable mic if browser lacks support
|
||||
if (!SpeechRecognition) {
|
||||
console.warn('Web Speech API not supported — mic disabled');
|
||||
$mic.disabled = true;
|
||||
$mic.title = 'Speech recognition not supported in this browser';
|
||||
}
|
||||
|
||||
function startRecording() {
|
||||
if (!SpeechRecognition) return;
|
||||
if (recognition) {
|
||||
// stale instance — tear it down first
|
||||
try { recognition.abort(); } catch {}
|
||||
recognition = null;
|
||||
}
|
||||
|
||||
recognition = new SpeechRecognition();
|
||||
recognition.continuous = false;
|
||||
recognition.interimResults = true;
|
||||
recognition.lang = 'en-US';
|
||||
|
||||
recognition.addEventListener('result', (e) => {
|
||||
// Build transcript from all results (interim + final)
|
||||
let transcript = '';
|
||||
for (let i = 0; i < e.results.length; i++) {
|
||||
transcript += e.results[i][0].transcript;
|
||||
}
|
||||
$input.value = transcript;
|
||||
});
|
||||
|
||||
recognition.addEventListener('end', () => {
|
||||
recording = false;
|
||||
$mic.classList.remove('recording');
|
||||
recognition = null;
|
||||
|
||||
// Auto-send if we got text
|
||||
const text = $input.value.trim();
|
||||
if (text) {
|
||||
$form.dispatchEvent(new Event('submit', { cancelable: true }));
|
||||
} else {
|
||||
flashMicEmpty();
|
||||
}
|
||||
});
|
||||
|
||||
recognition.addEventListener('error', (e) => {
|
||||
recording = false;
|
||||
$mic.classList.remove('recording');
|
||||
recognition = null;
|
||||
|
||||
// 'no-speech' and 'aborted' are expected — not worth alarming the user
|
||||
if (e.error === 'no-speech') {
|
||||
flashMicEmpty();
|
||||
} else if (e.error !== 'aborted') {
|
||||
console.warn('speech recognition error:', e.error);
|
||||
flashMicEmpty();
|
||||
}
|
||||
});
|
||||
|
||||
try {
|
||||
const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
|
||||
recordedChunks = [];
|
||||
// Prefer webm/opus if browser supports it (Chrome/FF). Safari may need fallback.
|
||||
const mimeTypes = ['audio/webm;codecs=opus', 'audio/webm', 'audio/mp4', ''];
|
||||
const mime = mimeTypes.find(m => !m || MediaRecorder.isTypeSupported(m)) || '';
|
||||
mediaRecorder = new MediaRecorder(stream, mime ? { mimeType: mime } : undefined);
|
||||
mediaRecorder.addEventListener('dataavailable', (e) => {
|
||||
if (e.data && e.data.size > 0) recordedChunks.push(e.data);
|
||||
});
|
||||
mediaRecorder.addEventListener('stop', () => {
|
||||
// release the mic immediately
|
||||
stream.getTracks().forEach(t => t.stop());
|
||||
handleRecorded();
|
||||
});
|
||||
mediaRecorder.start();
|
||||
recognition.start();
|
||||
recording = true;
|
||||
$mic.classList.add('recording');
|
||||
} catch (err) {
|
||||
console.warn('microphone unavailable:', err.message);
|
||||
console.warn('speech recognition failed to start:', err.message);
|
||||
recording = false;
|
||||
}
|
||||
}
|
||||
|
||||
function stopRecording() {
|
||||
if (!mediaRecorder || mediaRecorder.state === 'inactive') return;
|
||||
mediaRecorder.stop();
|
||||
recording = false;
|
||||
$mic.classList.remove('recording');
|
||||
$mic.classList.add('transcribing');
|
||||
}
|
||||
|
||||
async function handleRecorded() {
|
||||
if (!recordedChunks.length) {
|
||||
$mic.classList.remove('transcribing');
|
||||
return;
|
||||
}
|
||||
const blob = new Blob(recordedChunks, { type: recordedChunks[0].type || 'audio/webm' });
|
||||
const ext = (blob.type.split('/')[1] || 'webm').split(';')[0];
|
||||
const form = new FormData();
|
||||
form.append('audio', blob, `speech.${ext}`);
|
||||
|
||||
try {
|
||||
const resp = await fetch('/api/transcribe', { method: 'POST', body: form });
|
||||
if (!resp.ok) throw new Error(`HTTP ${resp.status}`);
|
||||
const { text } = await resp.json();
|
||||
if (text && text.trim()) {
|
||||
$input.value = text;
|
||||
// auto-send — Her vibe is "speak and she hears"
|
||||
$form.dispatchEvent(new Event('submit', { cancelable: true }));
|
||||
} else {
|
||||
flashMicEmpty();
|
||||
}
|
||||
} catch (err) {
|
||||
console.warn('transcription failed:', err.message);
|
||||
flashMicEmpty();
|
||||
} finally {
|
||||
$mic.classList.remove('transcribing');
|
||||
}
|
||||
if (!recognition) return;
|
||||
try { recognition.stop(); } catch {}
|
||||
// 'end' event handler cleans up recording state and sends
|
||||
}
|
||||
|
||||
function flashMicEmpty() {
|
||||
// brief visual hint that nothing was heard — no toast, no popup, just a flash
|
||||
$mic.classList.add('empty');
|
||||
setTimeout(() => $mic.classList.remove('empty'), 700);
|
||||
}
|
||||
|
||||
+26
-72
@@ -1,101 +1,55 @@
|
||||
"""Piper TTS adapter for chat.saiden.dev.
|
||||
"""HTTP TTS adapter for chat.saiden.dev — madcat-tts daemon (chatterbox).
|
||||
|
||||
Synthesises text → WAV bytes by subprocess'ing the `piper` CLI binary
|
||||
(already installed on every host that runs marauder-os).
|
||||
Calls the madcat-tts daemon's OpenAI-compatible /v1/audio/speech endpoint
|
||||
to synthesize text → WAV bytes via chatterbox voice cloning.
|
||||
|
||||
Designed to fail silently — if piper is missing or synthesis errors,
|
||||
Designed to fail silently — if the daemon is down or synthesis errors,
|
||||
the chat still works, just without voice.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
import logging
|
||||
import os
|
||||
import shutil
|
||||
import tempfile
|
||||
from pathlib import Path
|
||||
|
||||
import httpx
|
||||
|
||||
log = logging.getLogger("chat-saiden.tts")
|
||||
|
||||
# Where the voice .onnx files live across hosts.
|
||||
# Order: env override → macOS marauder → linux marauder → linux marauder-agent (mesh node) → linux ~/.local
|
||||
_VOICE_SEARCH_PATHS = [
|
||||
Path.home() / "Library/Application Support/marauder/voices",
|
||||
Path("/home") / os.environ.get("USER", "marauder") / ".local/share/marauder/voices",
|
||||
Path.home() / ".local/share/marauder/voices",
|
||||
Path.home() / ".local/share/psn/voices",
|
||||
Path.home() / ".local/share/piper/voices",
|
||||
]
|
||||
|
||||
|
||||
def _resolve_voice_path(name: str) -> Path | None:
|
||||
"""Return absolute path to a voice model by short name, or None."""
|
||||
# explicit override
|
||||
override = os.environ.get("TTS_VOICE_PATH")
|
||||
if override:
|
||||
p = Path(override)
|
||||
return p if p.exists() else None
|
||||
|
||||
for base in _VOICE_SEARCH_PATHS:
|
||||
candidate = base / f"{name}.onnx"
|
||||
if candidate.exists():
|
||||
return candidate
|
||||
return None
|
||||
|
||||
|
||||
PIPER_BIN = shutil.which("piper") or os.environ.get("PIPER_BIN")
|
||||
MADCAT_TTS_URL = os.environ.get("MADCAT_TTS_URL", "http://localhost:14099")
|
||||
|
||||
|
||||
class TTS:
|
||||
"""Subprocess-based piper synthesizer with graceful fallback."""
|
||||
"""HTTP-based madcat-tts synthesizer with graceful fallback."""
|
||||
|
||||
def __init__(self, voice: str = "en_US-amy-medium") -> None:
|
||||
def __init__(self, voice: str = "bt7274-en") -> None:
|
||||
self.voice = voice
|
||||
self.voice_path = _resolve_voice_path(voice)
|
||||
self.bin = PIPER_BIN
|
||||
if not self.bin:
|
||||
log.warning("piper binary not found on PATH — TTS disabled")
|
||||
elif not self.voice_path:
|
||||
log.warning("voice '%s' not found in known locations — TTS disabled", voice)
|
||||
else:
|
||||
log.info("TTS enabled — voice=%s path=%s bin=%s", voice, self.voice_path, self.bin)
|
||||
self._url = f"{MADCAT_TTS_URL.rstrip('/')}/v1/audio/speech"
|
||||
log.info("TTS enabled — voice=%s url=%s", voice, self._url)
|
||||
|
||||
@property
|
||||
def available(self) -> bool:
|
||||
return bool(self.bin and self.voice_path)
|
||||
return True
|
||||
|
||||
async def synthesize(self, text: str) -> bytes | None:
|
||||
"""Return WAV bytes, or None on failure / unavailable."""
|
||||
if not self.available:
|
||||
if not text or not text.strip():
|
||||
return None
|
||||
if not text.strip():
|
||||
return None
|
||||
|
||||
# piper wants an output file path (no stdout streaming for WAV in older versions)
|
||||
out = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
|
||||
out.close()
|
||||
out_path = out.name
|
||||
try:
|
||||
proc = await asyncio.create_subprocess_exec(
|
||||
self.bin,
|
||||
"--model", str(self.voice_path),
|
||||
"--output_file", out_path,
|
||||
stdin=asyncio.subprocess.PIPE,
|
||||
stdout=asyncio.subprocess.DEVNULL,
|
||||
stderr=asyncio.subprocess.PIPE,
|
||||
async with httpx.AsyncClient(timeout=15.0) as client:
|
||||
resp = await client.post(
|
||||
self._url,
|
||||
json={
|
||||
"input": text,
|
||||
"voice": self.voice,
|
||||
"response_format": "wav",
|
||||
},
|
||||
)
|
||||
_, stderr = await proc.communicate(text.encode("utf-8"))
|
||||
if proc.returncode != 0:
|
||||
log.error("piper exited %s: %s", proc.returncode, stderr.decode("utf-8", "replace")[:300])
|
||||
resp.raise_for_status()
|
||||
return resp.content
|
||||
except httpx.TimeoutException:
|
||||
log.warning("TTS timeout for voice=%s (text=%s…)", self.voice, text[:60])
|
||||
return None
|
||||
with open(out_path, "rb") as f:
|
||||
return f.read()
|
||||
except Exception:
|
||||
log.exception("piper synthesis failed")
|
||||
log.exception("TTS synthesis failed for voice=%s", self.voice)
|
||||
return None
|
||||
finally:
|
||||
try:
|
||||
os.unlink(out_path)
|
||||
except OSError:
|
||||
pass
|
||||
|
||||
Reference in New Issue
Block a user