scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-05 02:41:56 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 49

fix(ovhcloud): preserve segments/words/language in verbose_json transcription response

Browse source 
Fixes #25633

When  is used with the OVHCloud Whisper endpoint,
the provider returns a full verbose response containing , ,
, , etc. The previous implementation only extracted the
 field, silently dropping all other data.

This patch mirrors the approach used by the OpenAI Whisper implementation:
pass the full response JSON to  so
all OpenAI-compatible fields are preserved.

A normalisation step handles the case where OVHCloud returns 
instead of  to maintain backwards compatibility.
This commit is contained in:
Anai-Guo 2026-04-13 12:46:55 -07:00
parent d319cd8cc6
commit 668350a4b5
1 changed files with 18 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
litellm/llms/ovhcloud/audio_transcription/transformation.py
View file

@ -143,6 +143,11 @@ class OVHCloudAudioTranscriptionConfig(BaseAudioTranscriptionConfig):
) -> TranscriptionResponse:
"""
Transform OVHCloud audio transcription response to OpenAI-compatible TranscriptionResponse.
OVHCloud returns the full verbose_json format (including `segments`, `words`,
`language`, `duration`, etc.) when `response_format=verbose_json` is requested.
We pass the entire response JSON to TranscriptionResponse so all fields are
preserved rather than discarding everything except `text`.
"""
try:
response_json = raw_response.json()
@ -153,8 +158,19 @@ class OVHCloudAudioTranscriptionConfig(BaseAudioTranscriptionConfig):
headers=raw_response.headers,
)
text = response_json.get("text") or response_json.get("transcript") or ""
response = TranscriptionResponse(text=text)
# Normalise: OVHCloud may return `transcript` instead of `text`
if "transcript" in response_json and "text" not in response_json:
response_json["text"] = response_json.pop("transcript")
# Build a TranscriptionResponse that preserves all OpenAI-compatible fields
# (text, segments, words, language, duration, …) returned by the provider.
if any(
key in response_json
for key in TranscriptionResponse.model_fields.keys()
):
response = TranscriptionResponse(**response_json)
else:
response = TranscriptionResponse(text=response_json.get("text", ""))
response._hidden_params = response_json
return response