[Frontend] Add automatic language detection for Whisper transcription (#34342)

Signed-off-by: space_check <roman.vuskov@rwth-aachen.de>
Signed-off-by: Roman <45857014+spacecheck@users.noreply.github.com>
Co-authored-by: Nicolò Lucchesi <nicolo.lucchesi@gmail.com>
Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com>
This commit is contained in:
Roman
2026-02-21 04:49:41 -08:00
committed by GitHub
co-authored by Nicolò Lucchesi Nicolò Lucchesi
parent 272b535ab3
commit 98b0205c3c
5 changed files with 249 additions and 10 deletions
@@ -273,3 +273,30 @@ async def test_audio_with_max_tokens(whisper_client, mary_had_lamb):
out_text = out["text"]
out_tokens = tok(out_text, add_special_tokens=False)["input_ids"]
assert len(out_tokens) < 450 # ~Whisper max output len
@pytest.mark.asyncio
@pytest.mark.parametrize(
("fixture_name", "expected_lang", "expected_text"),
[
("mary_had_lamb", "en", ["Mary had a little lamb"]),
("foscolo", "it", ["zacinto", "sacre"]),
],
ids=["english", "italian"],
)
async def test_language_auto_detect(
whisper_client, fixture_name, expected_lang, expected_text, request
):
"""Auto-detect language when no language param is provided."""
audio_file = request.getfixturevalue(fixture_name)
transcription = await whisper_client.audio.transcriptions.create(
model=MODEL_NAME,
file=audio_file,
response_format="verbose_json",
temperature=0.0,
)
assert transcription.language == expected_lang
text_lower = transcription.text.lower()
assert any(word.lower() in text_lower for word in expected_text), (
f"Expected {expected_lang} text but got: {transcription.text}"
)