Fix: accept scalar audio formats in Voxtral transcription requests

[Sunt-ing]

What does this PR do?

VoxtralProcessor.apply_transcription_request() documents format as either a string or a list. The array-audio path did not normalize a scalar string, so format="wav" was treated as three per-sample formats and a single local audio array failed before the transcription request was encoded.

This broadcasts a scalar format to the audio batch length, keeps the existing length check for per-sample format lists, and raises a clear optional-dependency error before encoding array audio without soundfile.

Repro and output

import numpy as np
from transformers import AutoProcessor


model_id = "mistralai/Voxtral-Mini-3B-2507"
processor = AutoProcessor.from_pretrained(model_id)
audio = np.zeros(16000, dtype=np.float32)

for fmt in ["wav", ["wav"]]:
    print("CASE", fmt)
    try:
        out = processor.apply_transcription_request(
            audio=audio,
            model_id=model_id,
            format=fmt,
            sampling_rate=16000,
        )
        print("OK", sorted(out.keys()), out["input_features"].shape, out["input_ids"].shape)
    except Exception as exc:
        print("ERR", type(exc).__name__, str(exc))

Before this PR:

CASE wav
ERR ValueError When passed as a list of audio, the length (1) must match the number of format (3)
CASE ['wav']
OK ['attention_mask', 'input_features', 'input_ids'] torch.Size([1, 128, 3000]) torch.Size([1, 380])

After this PR:

CASE wav
OK ['attention_mask', 'input_features', 'input_ids'] torch.Size([1, 128, 3000]) torch.Size([1, 380])
CASE ['wav']
OK ['attention_mask', 'input_features', 'input_ids'] torch.Size([1, 128, 3000]) torch.Size([1, 380])

Tests

RUN_SLOW=1 python -m pytest -q -s tests/models/voxtral/test_modeling_voxtral.py::VoxtralForConditionalGenerationIntegrationTest::test_transcription_request_accepts_single_format_for_array_audio

tests/models/voxtral/test_modeling_voxtral.py::VoxtralForConditionalGenerationIntegrationTest::test_transcription_request_accepts_single_format_for_array_audio PASSED

======================= 1 passed, 11 warnings in 26.85s ========================

python -m ruff check src/transformers/models/voxtral/processing_voxtral.py tests/models/voxtral/test_modeling_voxtral.py
All checks passed!

git diff --check

Code Agent Policy

• I confirm that this is not a pure code agent PR.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline and the
  Pull Request checks?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes according to the guidelines?
• Did you write any new necessary tests?

Who can review?

cc @zucchini-nlp @Rocketknight1

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: voxtral

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 28673130375
Result: success | Grafana metrics are not available yet.

[Rocketknight1]

LGTM but cc @eustlb @ebezzam for audio to confirm!

[eustlb]

LGTM, thanks a lot !! 🤗

[eustlb]

To be replace with requires_backend when #47113 is merged (other fixed there is this get merged before)

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.