Skip to content

fix(tts): transcode xAI TTS output to Opus when .ogg requested (#57213) #57518

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
AlexFucuson9 wants to merge 1 commit into from
Closed

fix(tts): transcode xAI TTS output to Opus when .ogg requested (#57213) #57518

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 32 additions & 1 deletion tools/tts_tool.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1229,7 +1229,18 @@ def _generate_xai_tts(text: str, output_path: str, tts_config: Dict[str, Any]) -

# Match the documented minimal POST /v1/tts shape by default. Only send
# output_format when Hermes actually needs a non-default format/override.
# The xAI /v1/tts API only supports mp3 and wav codecs — it has no native
# Opus/OGG output. When the caller requests an .ogg file (e.g. Telegram
# voice bubbles via the auto-reply path), we synthesize to a temp .mp3
# first and transcode with ffmpeg after the API call returns. Without
# this, MP3 bytes are silently written into the .ogg-named file, producing
# broken voice bubbles (issue #57213).
_needs_opus_transcode = output_path.endswith(".ogg")
codec = "wav" if output_path.endswith(".wav") else "mp3"
# When transcoding to opus, write the API response to a temp mp3 file so
# the final .ogg lands at the caller's requested output_path.
_mp3_tmp: str = output_path.rsplit(".", 1)[0] + "._xaitemp.mp3" if _needs_opus_transcode else ""
_write_path: str = _mp3_tmp if _needs_opus_transcode else output_path
payload: Dict[str, Any] = {
"text": text,
"voice_id": voice_id,
Expand Down Expand Up @@ -1271,9 +1282,29 @@ def _generate_xai_tts(text: str, output_path: str, tts_config: Dict[str, Any]) -
)
response.raise_for_status()

with open(output_path, "wb") as f:
with open(_write_path, "wb") as f:
f.write(response.content)

# Transcode to Opus/OGG if the caller requested an .ogg output path.
# xAI's /v1/tts API has no native Opus codec, so we write mp3 to a temp
# file and convert with ffmpeg (same pattern as the dispatcher-level
# _convert_to_opus used by other providers).
if _needs_opus_transcode:
opus_path = _convert_to_opus(_mp3_tmp)
if opus_path:
# Clean up the temp mp3 file
try:
os.unlink(_mp3_tmp)
except OSError:
pass
return opus_path

@teknium1 teknium1 Jul 15, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

_convert_to_opus() derives an OGG beside this temporary MP3, but text_to_speech_tool() ignores the xAI generator's return value (tools/tts_tool.py:2443-2445) and still validates the original requested .ogg path. Move the converted file to output_path before returning, or propagate the returned path through the dispatcher.

# ffmpeg not available or conversion failed — fall back to the mp3
# file so the caller at least gets *some* audio rather than nothing.
logger.warning(
"xAI TTS: ffmpeg opus conversion failed; returning mp3 at %s", _mp3_tmp
)
return _mp3_tmp

return output_path


Expand Down