README.md

metadata

annotations_creators:
  - expert-generated
language_creators:
  - expert-generated
language:
  - en
license:
  - mit
multilinguality:
  - monolingual
pretty_name: VoiceCodeBench
size_categories:
  - n<1K
source_datasets: []
task_categories:
  - automatic-speech-recognition
tags:
  - audio
  - speech
  - speech-recognition
  - speech-to-text
  - automatic-speech-recognition
  - asr
  - asr-benchmark
  - benchmark
  - evaluation
  - structured-token-recovery
  - entity-recovery
  - workplace-speech
  - english
dataset_info:
  - config_name: default
    features:
      - name: file_name
        dtype: string
      - name: audio_id
        dtype: string
      - name: language
        dtype: string
      - name: duration
        dtype: float64
      - name: domain
        dtype: string
      - name: scenario
        dtype: string
      - name: difficulty
        dtype: string
      - name: speaker
        struct:
          - name: id
            dtype: string
          - name: sex
            dtype: string
          - name: accent
            dtype: string
          - name: age_bucket
            dtype: string
      - name: audio_quality
        struct:
          - name: snr_db
            dtype: float64
          - name: noise_rms_dbfs
            dtype: float64
          - name: speech_rms_dbfs
            dtype: float64
          - name: loudness_lufs
            dtype: float64
          - name: click_pop_count_per_min
            dtype: float64
      - name: transcripts
        struct:
          - name: template
            dtype: string
          - name: acoustic
            dtype: string
          - name: canonical
            dtype: string
      - name: entities
        list:
          - name: id
            dtype: string
          - name: type
            dtype: string
          - name: role
            dtype: string
          - name: acoustic
            dtype: string
          - name: canonical
            dtype: string
      - name: entity_types
        list: string
      - name: entity_count
        dtype: int64
    splits:
      - name: test
        num_bytes: 911121
        num_examples: 300
    download_size: 1068651
    dataset_size: 911121
configs:
  - config_name: default
    default: true
    data_files:
      - split: test
        path: data/metadata.jsonl

VoiceCodeBench

VoiceCodeBench is a test-only benchmark for evaluating whether automatic speech recognition (ASR) systems preserve exact structured values in English workplace speech.

Paper: VoiceCodeBench: Evaluating Exact Structured-Token Recovery in Automatic Speech Recognition

The benchmark targets cases where a transcript is software input: callback numbers, email addresses, command-line flags, file paths, URLs, account identifiers, dates, measurements, and similar values that downstream systems may parse, route, store, compare, or execute.

Contents

• 300 human-recorded English WAV segments, totaling 5.587 hours.
• 85 anonymized speaker IDs.
• 1,482 audited target entities across 26 structured entity types.
• 8 workplace workflow domains.
• 15 tracked baseline ASR system outputs.

Released files:

• data/audio/*.wav: benchmark audio files.
• data/metadata.jsonl: transcripts, entity annotations, speaker metadata, and audio-quality metadata.
• baselines/predictions/*.json: baseline transcripts plus entity-match decisions.
• baselines/results.csv: aggregate baseline table.
• scripts/: transcription, entity verification, scoring, and figure commands.
• paper/: paper source and PDF.
• DATASET_CARD.md: datasheet-style documentation for motivation, composition, collection, consent, intended use, limits, maintenance, and licensing.

Task

Each item contains an audio recording, three transcript layers, and target entities:

• template: script text with entity placeholders.
• acoustic: what the speaker is expected to say aloud.
• canonical: the written value a downstream application needs.

For example, "double dash dry dash run" maps to --dry-run, and "all caps database underscore URL" maps to DATABASE_URL.

ASR systems are evaluated under a raw-audio-only protocol. The system receives only the audio file; benchmark-specific prompts, target entity lists, domain labels, custom vocabulary, grammar constraints, candidate values, and post-ASR correction are excluded from the main setting.

Metrics

VoiceCodeBench reports WER as a broad transcript-quality diagnostic, but its main entity-sensitive metrics are:

CTEM = correct target entities / target entities
TSR  = recordings with all target entities correct / recordings

Canonical Token/Entity Match (CTEM) measures value-level recovery. Task Success Rate (TSR) measures whether every target entity in a recording was recovered.

Baselines

The tracked baseline suite contains 15 ASR systems across batch and streaming modes. Current aggregate ranges:

• WER: 8.6% to 26.1%.
• CTEM: 72.5% to 91.6%.
• TSR: 25.7% to 68.7%.

The strongest baseline by TSR is deepgram_nova3 at 68.7%. The strongest baseline by CTEM is elevenlabs_scribe_v2 at 91.6%. modal_inkling scores 24.3% WER, 84.3% CTEM, and 49.7% TSR in batch mode. modal_nvidia_parakeet_tdt_0_6b_v3 scores 23.7% WER, 78.6% CTEM, and 37.3% TSR in batch mode. modal_meta_omniasr_llm_unlimited_7b_v2 scores 26.1% WER, 72.5% CTEM, and 25.7% TSR in batch mode.

Install

For scoring released baseline artifacts:

python -m pip install -e .

Optional extras are available for heavier workflows:

python -m pip install -e ".[providers]"  # run new ASR baselines
python -m pip install -e ".[figures]"    # regenerate paper figures
python -m pip install -e ".[dev]"        # provider and figure dependencies

External tools are only needed for optional workflows:

• ffmpeg for live provider audio conversion, including streaming PCM and Inkling's 16 kHz WAV input.
• gcloud for Google Cloud transcription when application-default credentials are not already configured.
• latexmk for rebuilding the paper PDF.

Tests

python -m pytest

Reproduce

./scripts/reproduce_release.sh

This creates .venv if needed, installs the package in editable mode with the figures extra, validates metadata, scores the released baseline transcripts/entity matches, rewrites baselines/results.csv, and regenerates paper/figures/wer_entity_scatter.pdf from the frozen 12-model paper baseline set. Post-publication baselines such as Inkling and Parakeet are excluded from that figure.

To generate the current benchmark figure with every row in baselines/results.csv, run:

vcb-make-figures --model-set all

This writes baselines/figures/wer_entity_scatter.pdf by default. Use --model-set paper to regenerate only the frozen paper figure, or --output to choose another path.

To rebuild the paper PDF:

cd paper
latexmk -pdf -interaction=nonstopmode -halt-on-error voice-code-bench.tex

Experimental New Baselines

The released scoring and reproduction commands are the stable script surface. Live provider runs are included to make the tracked baselines auditable, but provider APIs and websocket protocols change over time.

Running new ASR baselines requires provider credentials. Copy scripts/.secret.example to a private secret file or set equivalent environment variables, including the Modal endpoint settings described below, then run:

python -m venv .venv
. .venv/bin/activate
python -m pip install -e ".[providers]"

vcb-run \
  --stt-mode all \
  --output-dir runs/full-local

Modal Inkling

Thinking Machines Lab Inkling is available as the batch model ID modal_inkling. Modal exposes thinkingmachines/Inkling-NVFP4 through a shared, OpenAI-compatible managed endpoint with token-based pricing; no dedicated model deployment is required. Open the Inkling endpoint page, create a Modal account if needed, create the managed endpoint, and copy its URL. Create a proxy token and configure the private scripts/.secret file with the endpoint URL, token ID, and token secret:

MODAL_INKLING_ENDPOINT=https://<your-endpoint-host>
MODAL_PROXY_TOKEN_ID=wk-...
MODAL_PROXY_TOKEN_SECRET=ws-...

The adapter accepts the base endpoint URL, /v1 URL, or full /v1/chat/completions URL shown by Modal. Check the endpoint page for current pricing and rate limits before a full benchmark run. Never commit scripts/.secret; it is ignored by Git.

To smoke-test one recording after configuring real credentials:

vcb-transcribe \
  --stt-model-ids modal_inkling \
  --limit 1 \
  --output-dir runs/inkling-smoke

To run the full transcription, entity-verification, and scoring pipeline, also configure OPENAI_API_KEY for the entity verifier and run:

vcb-run \
  --stt-model-ids modal_inkling \
  --output-dir runs/inkling-full

The adapter converts source recordings to the model's documented 16 kHz mono WAV input, sends the audio through Modal's chat-completions audio_url format, uses Thinking Machines' documented transcription prompt and text-before-audio message order, and sets reasoning effort to max (0.99). The transcript response is stored verbatim without benchmark-specific hints or post-ASR correction.

Thinking Machines' limited-time free Inkling Playground is useful for manually spot-checking audio, but it is a chat interface rather than the reproducible batch API used by this benchmark.

Modal NVIDIA Parakeet TDT 0.6B v3

NVIDIA Parakeet TDT 0.6B v3 is available as the batch model ID modal_nvidia_parakeet_tdt_0_6b_v3 through the checked-in Modal app modal/parakeet_tdt_0_6b_v3.py. It self-hosts nvidia/parakeet-tdt-0.6b-v3 on an L40S using official Hugging Face weights and the shared Modal-ASR JSON transport. Configure these private secrets:

MODAL_PARAKEET_ENDPOINT=https://<your-parakeet-web-function>.modal.run
MODAL_PROXY_TOKEN_ID=wk-...
MODAL_PROXY_TOKEN_SECRET=ws-...

Warm and deploy the app as described in scripts/MODAL.md, then smoke-test the longest recording with a one-row metadata file rather than --limit 1. The full run command is:

vcb-run \
  --stt-model-ids modal_nvidia_parakeet_tdt_0_6b_v3 \
  --resume \
  --output-dir runs/parakeet-full

The adapter converts each source recording to one complete 16 kHz mono WAV and sends only raw audio bytes. No chunking, prompts, custom vocabulary, target-value hints, canonicalization, or post-ASR correction are used; the benchmark stores only the returned transcript verbatim.

Modal Meta OmniASR LLM Unlimited 7B v2

Meta OmniASR LLM Unlimited 7B v2 is registered as the batch model ID modal_meta_omniasr_llm_unlimited_7b_v2 through the checked-in Modal app modal/omniasr_llm_unlimited_7b_v2.py. It self-hosts omniASR_LLM_Unlimited_7B_v2 on an L40S with omnilingual-asr==0.2.0, a persistent fairseq2 asset cache, and the shared Modal-ASR JSON transport. Configure these private secrets:

MODAL_OMNIASR_ENDPOINT=https://<your-omniasr-web-function>.modal.run
MODAL_PROXY_TOKEN_ID=wk-...
MODAL_PROXY_TOKEN_SECRET=ws-...

Warm and deploy the app as described in scripts/MODAL.md, then smoke-test the longest recording, education_workplace_006 (data/audio/261.wav, 122.875 seconds), with a one-row metadata file rather than --limit 1. The full run command is:

vcb-run \
  --stt-model-ids modal_meta_omniasr_llm_unlimited_7b_v2 \
  --resume \
  --output-dir runs/omniasr-full

The adapter converts each source recording to one complete 16 kHz mono WAV, sends language="eng_Latn", and stores only the returned transcript verbatim. No chunking, prompts, custom vocabulary, target-value hints, canonicalization, or post-ASR correction are used. A publishable OmniASR baseline additionally requires a real deployed endpoint, proxy token, entity-verifier access or cache, a complete 300-recording artifact, recomputed scores, and the current all-model figure; mock output must not be copied into baselines/predictions/.

For publishable results, report provider, model name, API endpoint or endpoint family, evaluation date, mode, inference settings, and any prompting, custom vocabulary, post-processing, fine-tuning, or canonicalization.

Use And Limits

VoiceCodeBench is intended for diagnostic ASR evaluation, provider comparison, regression tracking, and per-entity risk analysis. It is not intended as a training corpus, hidden leaderboard, universal ASR-quality measure, speaker identification resource, biometric dataset, or demographic profiling dataset.

See DATASET_CARD.md for the full dataset statement.

Citation

@misc{voicecodebench2026,
  title = {VoiceCodeBench: Evaluating Exact Structured-Token Recovery in Automatic Speech Recognition},
  author = {Baumgartner, Tyler and Tai, Brandon and Kaelin-Martin, Lisa and Fan, Candice and Debaupte, Luc and Wang, Bill and Zhong, Yi},
  year = {2026},
  note = {Benchmark dataset and paper}
}

License

VoiceCodeBench is released under the MIT License. See LICENSE.