Infer audio encoding format from source instead of defaulting to MP3

[jaredoconnell]

Summary

Uses the source's audio format rather than defaulting to MP3

Details

• If no format is provided by the user, and it cannot be inferred, it defaults to WAV and warns the user.
• Simply uses the format read from the dataset if none is provided.
• Includes a fix for a missing super call in the response handler.

Test Plan

Make sure you install vLLM with the audio features.

You can run this with multiple datasets. Some options include:

guidellm benchmark run   --target "http://localhost:8000"   --request-type /v1/audio/transcriptions   --profile kind=synchronous   --max-requests 10   --data '{"kind": "huggingface", "source": "google/fleurs", "load_kwargs": {"name": "en_us", "split": "test"}}'   --data-column-mapper '{"column_mappings": {"audio_column": "audio"}}'   --disable-progress

guidellm benchmark run   --target "http://localhost:8000"   --request-type /v1/audio/transcriptions   --profile kind=synchronous   --max-requests 10   --data '{"kind": "huggingface", "source": "openslr/librispeech_asr", "load_kwargs": {"name": "clean", "split": "test"}}'   --data-column-mapper '{"column_mappings": {"audio_column": "audio"}}'   --disable-progress

To explicitly use MP3 like it used to, do:

guidellm benchmark run \
  --target "http://localhost:8000" \
  --request-type /v1/audio/transcriptions \
  --profile kind=synchronous \
  --max-requests 10 \
  --data '{"kind": "huggingface", "source": "openslr/librispeech_asr", "load_kwargs": {"name": "clean", "split": "test"}}' \
  --data-column-mapper '{"column_mappings": {"audio_column": "audio"}}' \
  --data-preprocessors '{"kind": "encode_media", "audio_kwargs": {"audio_format": "mp3"}}' \
  --disable-progress

Related Issues

• Resolves Avoid default MP3 transcoding for transcription benchmarks #623

---

• "I certify that all code in this PR is my own, except as noted below."

Use of AI

• Includes code generated or substantially modified by an AI agent
• Includes tests generated or substantially modified by an AI agent

NOTE: the Generated-by or Assisted-by trailers should be used in git commit messages when code or tests were generated or substantially modified by an AI agent, as described in the project's DEVELOPING.md file.

---

git log

commit 6cc6e9f
Author: Jared O'Connell joconnel@redhat.com
Date: Thu Jun 11 15:00:08 2026 -0400

Infer audio encoding format from source instead of defaulting to MP3

Generated-by: Cursor AI Claude Opus 4.6
Signed-off-by: Jared O'Connell <joconnel@redhat.com>

---

Generated-by: Cursor AI Claude Opus 4.6
Signed-off-by: Jared O'Connell joconnel@redhat.com