Files
hermes-agent/tools/voice_mode_transcript.py

162 lines
6.3 KiB
Python

"""Pure-text helpers for voice mode: Whisper hallucination filter, voice-chat
stop phrases, and the TTS self-echo guard. No audio dependencies."""
import difflib
import re
from typing import Optional
def _voice_config() -> dict:
"""``voice`` section of config.yaml, or ``{}`` when missing, malformed,
or the config system can't be imported (broken config mid-install)."""
try:
from hermes_cli.config import load_config
voice_cfg = load_config().get("voice", {})
return voice_cfg if isinstance(voice_cfg, dict) else {}
except Exception:
return {}
# Whisper commonly hallucinates these phrases on silent/near-silent audio
# (matched with trailing '.'/'!' stripped, so the bare form suffices).
WHISPER_HALLUCINATIONS = {
"thank you", "thanks for watching", "subscribe to my channel", "like and subscribe",
"please subscribe", "thank you for watching", "bye", "you", "the end",
# Non-English hallucinations (common on silence)
"продолжение следует", "sous-titres", "sous-titres réalisés par la communauté d'amara.org",
"sottotitoli creati dalla comunità amara.org", "untertitel von stephanie geiges",
"amara.org", "www.mooji.org", "ご視聴ありがとうございました",
}
# Repetitive hallucinations (e.g. "Thank you. Thank you. Thank you.")
_HALLUCINATION_REPEAT_RE = re.compile(
r'^(?:thank you|thanks|bye|you|ok|okay|the end|\.|\s|,|!)+$',
flags=re.IGNORECASE,
)
def is_whisper_hallucination(transcript: str) -> bool:
"""Check if a transcript is a known Whisper hallucination on silence."""
cleaned = transcript.strip().lower()
if not cleaned:
return True
return (
cleaned.rstrip('.!') in WHISPER_HALLUCINATIONS
or bool(_HALLUCINATION_REPEAT_RE.match(cleaned))
)
DEFAULT_VOICE_STOP_PHRASES = ("stop",)
def _load_voice_stop_phrases() -> tuple:
"""Configured ``voice.stop_phrases`` (default ``("stop",)``); an empty tuple
disables the feature. Malformed config (dict, list of non-strings) falls
back to the default rather than crashing the voice loop."""
try:
raw = _voice_config().get("stop_phrases", DEFAULT_VOICE_STOP_PHRASES)
if isinstance(raw, str):
raw = [raw]
if isinstance(raw, (list, tuple)):
return tuple(
str(p).strip().lower() for p in raw
if isinstance(p, (str, int, float)) and str(p).strip()
)
except Exception:
pass
return DEFAULT_VOICE_STOP_PHRASES
def _configured_stop_phrases() -> tuple:
"""Resolve ``_load_voice_stop_phrases`` through ``tools.voice_mode`` so
``patch("tools.voice_mode._load_voice_stop_phrases")`` still takes effect."""
from tools import voice_mode as _vm
return _vm._load_voice_stop_phrases()
def is_voice_stop_phrase(transcript: str, stop_phrases: Optional[tuple] = None) -> bool:
"""True when *transcript* is EXACTLY a configured stop phrase.
Deliberately strict: the whole utterance — lowercased, surrounding
punctuation stripped — must equal a phrase, so "stop doing that and try
again" still reaches the agent. ``voice.stop_phrases: []`` disables.
"""
if not transcript:
return False
cleaned = transcript.strip().lower().strip(".,!?;: \t\n\"'")
if not cleaned:
return False
if stop_phrases is None:
stop_phrases = _configured_stop_phrases()
return cleaned in stop_phrases
# Similarity ratio (difflib.SequenceMatcher) above which a playback-phase barge
# transcript is treated as a self-capture of Hermes' own TTS: the full-duplex
# listener has no echo cancellation, so speaker bleed can trip the barge
# trigger and get transcribed near-verbatim (a TTS -> STT -> TTS loop).
DEFAULT_TTS_ECHO_SIMILARITY_THRESHOLD = 0.6
# Minimum normalized-transcript length before the sliding-window fallback
# runs. Below this a genuine one-word barge-in ("yes") landing verbatim inside
# a longer reply would score a trivial 1.0 and be misread as self-capture; a
# real self-capture spans pre-roll plus time-to-silence, so it is longer.
MIN_FRAGMENT_LENGTH_FOR_ECHO = 10
def _normalize_for_echo_compare(text: str) -> str:
return re.sub(r"\s+", " ", text).strip().lower()
def is_tts_echo(
transcript: str,
spoken_text: str,
threshold: float = DEFAULT_TTS_ECHO_SIMILARITY_THRESHOLD,
) -> bool:
"""True when *transcript* looks like a self-capture of *spoken_text*.
Character-level similarity (language-agnostic, no word tokenization): a
genuine user interjection is very unlikely to closely match Hermes' own
words, so a high ratio signals speaker-bleed self-capture (fail-closed
guard for the playback-phase listener, which has no echo cancellation).
The playback-phase capture is cut when the trigger fires and only spans
pre-roll plus time-to-silence, so for replies longer than a clause the
transcript is a short FRAGMENT of `spoken_text` and the whole-string
ratio dilutes toward 0. When it misses, a window sized to the transcript
slides across `spoken_text` (character-based, so it works without word
boundaries). Transcripts shorter than `MIN_FRAGMENT_LENGTH_FOR_ECHO` skip
this fallback: a short interjection trivially matches a short window.
"""
if not transcript or not spoken_text:
return False
a = _normalize_for_echo_compare(transcript)
b = _normalize_for_echo_compare(spoken_text)
if not a or not b:
return False
def _similar(x: str, y: str) -> bool:
return difflib.SequenceMatcher(None, x, y).ratio() >= threshold
if _similar(a, b):
return True
if len(a) < MIN_FRAGMENT_LENGTH_FOR_ECHO or len(a) >= len(b):
return False
return any(_similar(a, b[start : start + len(a)]) for start in range(0, len(b) - len(a) + 1))
def voice_stop_hint() -> str:
"""One-line 'Say "stop" to end the voice chat.' hint for voice-mode start.
Uses the first ``voice.stop_phrases`` entry so a custom phrase renders
correctly; returns "" when stop phrases are disabled so surfaces show no
hint. Every surface announcing voice-mode start (CLI, TUI, desktop) uses
this one owner instead of hardcoding the wording.
"""
phrases = _configured_stop_phrases()
if not phrases:
return ""
return f'Say "{phrases[0]}" to end the voice chat.'