mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-30 19:09:28 +00:00
fae29c8410
Root-cause fix for the 'TTS voice bubble broken' issue family (#57048, #54589, #57213, #58845, #14841, #45557, #57049). Two class-level defects: 1. Several backends silently write MP3/WAV bytes into a .ogg output path (Edge only emits MP3, Piper writes WAV, xAI writes MP3, some OpenAI-compatible servers ignore response_format=opus). Platforms that need real Ogg/Opus render 0-second/broken voice bubbles. Instead of per-provider patches, text_to_speech_tool now sniffs magic bytes once after synthesis (_sniff_audio_container) and repairs the container centrally (_repair_ogg_container): ffmpeg transcode in place, or rename to the honest extension when ffmpeg is unavailable. Covers every current and future provider, including command providers and plugins. 2. want_opus only recognized Telegram, so Matrix/Feishu/WhatsApp/Signal auto-TTS voice replies were synthesized as MP3 and delivered as broken attachments. New OPUS_VOICE_PLATFORMS set covers all voice-bubble platforms. _convert_to_opus refactored onto a shared _ffmpeg_transcode_to_opus that supports safe in-place transcodes (-f ogg forced muxer, temp file + os.replace). Tests: tests/tools/test_tts_container_repair.py (13 tests incl. a live ffmpeg round-trip); full TTS suite green (153 + 187 passed).
136 lines
4.8 KiB
Python
136 lines
4.8 KiB
Python
"""Tests for the class-level TTS .ogg container repair.
|
|
|
|
Root cause class (#57048, #54589, #57213, #58845, #14841, #45557, #57049):
|
|
several TTS backends silently write MP3/WAV bytes into a ``.ogg`` output
|
|
path (Edge only emits MP3; Piper writes WAV; xAI writes MP3; some
|
|
OpenAI-compatible servers ignore ``response_format="opus"``). Platforms
|
|
that require real Ogg/Opus for native voice bubbles (Telegram, Matrix,
|
|
Feishu, WhatsApp, Signal) then render a broken 0-second bubble.
|
|
|
|
Instead of per-provider fixes, ``text_to_speech_tool`` sniffs the magic
|
|
bytes once after synthesis and repairs the container centrally.
|
|
"""
|
|
|
|
import struct
|
|
from unittest.mock import patch
|
|
|
|
import pytest
|
|
|
|
from tools.tts_tool import (
|
|
OPUS_VOICE_PLATFORMS,
|
|
_repair_ogg_container,
|
|
_sniff_audio_container,
|
|
)
|
|
|
|
MP3_ID3 = b"ID3\x04\x00\x00\x00\x00\x00\x00" + b"\x00" * 64
|
|
MP3_FRAME = b"\xff\xfb\x90\x00" + b"\x00" * 64
|
|
OGG = b"OggS\x00\x02" + b"\x00" * 64
|
|
FLAC = b"fLaC" + b"\x00" * 64
|
|
|
|
|
|
def _wav_bytes() -> bytes:
|
|
return b"RIFF" + struct.pack("<I", 36) + b"WAVE" + b"\x00" * 64
|
|
|
|
|
|
class TestSniffAudioContainer:
|
|
@pytest.mark.parametrize(
|
|
"data,expected",
|
|
[
|
|
(MP3_ID3, "mp3"),
|
|
(MP3_FRAME, "mp3"),
|
|
(OGG, "ogg"),
|
|
(FLAC, "flac"),
|
|
],
|
|
)
|
|
def test_magic_bytes(self, tmp_path, data, expected):
|
|
p = tmp_path / "a.bin"
|
|
p.write_bytes(data)
|
|
assert _sniff_audio_container(str(p)) == expected
|
|
|
|
def test_wav(self, tmp_path):
|
|
p = tmp_path / "a.bin"
|
|
p.write_bytes(_wav_bytes())
|
|
assert _sniff_audio_container(str(p)) == "wav"
|
|
|
|
def test_unknown_and_missing(self, tmp_path):
|
|
p = tmp_path / "a.bin"
|
|
p.write_bytes(b"\x00\x01\x02\x03" * 8)
|
|
assert _sniff_audio_container(str(p)) == "unknown"
|
|
assert _sniff_audio_container(str(tmp_path / "missing")) == "unknown"
|
|
|
|
|
|
class TestRepairOggContainer:
|
|
def test_real_ogg_untouched(self, tmp_path):
|
|
p = tmp_path / "v.ogg"
|
|
p.write_bytes(OGG)
|
|
assert _repair_ogg_container(str(p)) == str(p)
|
|
assert p.read_bytes() == OGG
|
|
|
|
def test_non_ogg_extension_untouched(self, tmp_path):
|
|
p = tmp_path / "v.mp3"
|
|
p.write_bytes(MP3_ID3)
|
|
assert _repair_ogg_container(str(p)) == str(p)
|
|
|
|
def test_mp3_in_ogg_transcoded(self, tmp_path):
|
|
p = tmp_path / "v.ogg"
|
|
p.write_bytes(MP3_ID3)
|
|
|
|
def fake_transcode(input_path, ogg_path):
|
|
# simulate in-place ffmpeg success
|
|
with open(ogg_path, "wb") as fh:
|
|
fh.write(OGG)
|
|
return ogg_path
|
|
|
|
with patch("tools.tts_tool._ffmpeg_transcode_to_opus", fake_transcode):
|
|
result = _repair_ogg_container(str(p))
|
|
|
|
assert result == str(p)
|
|
assert p.read_bytes()[:4] == b"OggS"
|
|
|
|
def test_wav_in_ogg_transcoded(self, tmp_path):
|
|
p = tmp_path / "v.ogg"
|
|
p.write_bytes(_wav_bytes())
|
|
with patch("tools.tts_tool._ffmpeg_transcode_to_opus",
|
|
lambda i, o: (open(o, "wb").write(OGG), o)[1]):
|
|
assert _repair_ogg_container(str(p)) == str(p)
|
|
assert p.read_bytes()[:4] == b"OggS"
|
|
|
|
def test_no_ffmpeg_renames_to_honest_extension(self, tmp_path):
|
|
p = tmp_path / "v.ogg"
|
|
p.write_bytes(MP3_FRAME)
|
|
with patch("tools.tts_tool._ffmpeg_transcode_to_opus", lambda i, o: None):
|
|
result = _repair_ogg_container(str(p))
|
|
assert result == str(tmp_path / "v.mp3")
|
|
assert not p.exists()
|
|
assert (tmp_path / "v.mp3").exists()
|
|
|
|
def test_ffmpeg_real_transcode_if_available(self, tmp_path):
|
|
"""Live ffmpeg round-trip when the binary exists (skipped otherwise)."""
|
|
import shutil as _shutil
|
|
import subprocess as _sp
|
|
|
|
if not _shutil.which("ffmpeg"):
|
|
pytest.skip("ffmpeg not installed")
|
|
# Synthesize a real tiny mp3 with ffmpeg, misname it .ogg
|
|
p = tmp_path / "v.ogg"
|
|
_sp.run(
|
|
["ffmpeg", "-f", "lavfi", "-i", "sine=frequency=440:duration=0.3",
|
|
"-acodec", "libmp3lame", "-f", "mp3", str(p), "-y"],
|
|
capture_output=True, check=True,
|
|
)
|
|
assert _sniff_audio_container(str(p)) == "mp3"
|
|
result = _repair_ogg_container(str(p))
|
|
assert result == str(p)
|
|
assert _sniff_audio_container(str(p)) == "ogg"
|
|
|
|
|
|
class TestOpusPlatformSet:
|
|
def test_opus_platforms_cover_voice_bubble_platforms(self):
|
|
# Behavior contract: the platforms whose adapters deliver native
|
|
# voice bubbles only for Ogg/Opus must be recognized.
|
|
for platform in ("telegram", "matrix", "feishu", "whatsapp", "signal"):
|
|
assert platform in OPUS_VOICE_PLATFORMS
|
|
|
|
def test_cli_not_included(self):
|
|
assert "" not in OPUS_VOICE_PLATFORMS
|
|
assert "cli" not in OPUS_VOICE_PLATFORMS
|