VERSA

SPEECH · AUDIO · MUSIC

Understand your audio.
From every angle.

Versatile Evaluation of Speech and Audio.
One open-source toolkit to bring quality, intelligibility, and acoustic analysis into your evaluation workflow.

Get started Read the paper

Built for researchers. Ready for your next experiment.

THE EVALUATION TOOLKIT01 / VERSA
YOUR AUDIO ↓
VERSAOne configurable workflow
Perceptual qualityIntelligibilityAcoustic propertiesAudio distributions
MULTIPLE PERSPECTIVES. A CLEARER PICTURE.
96Discoverable metrics
436GitHub stars
53GitHub forks
4Metric categories

Updated · Refreshed daily · Sources & how we count ↗

A WIDER PERSPECTIVE

Good audio has more
than one dimension.

Choose the metrics that fit your task, whether you have a reference recording, a text prompt, or a whole collection of audio.

01 / INDEPENDENT

Listen without a reference.

Assess an audio sample on its own, from perceptual quality to speaker and signal characteristics.

DNSMOSNISQAUTMOS
02 / DEPENDENT

Compare with the original.

Measure how predicted audio differs from a reference in fidelity, intelligibility, and acoustic detail.

PESQSTOISI-SDR
03 / NON-MATCH

Connect sound and context.

Evaluate with non-matching references or information from other modalities, including text.

Cross-modal evaluation
04 / DISTRIBUTIONAL

See the bigger picture.

Compare statistical properties across audio collections to evaluate a system beyond individual samples.

FADKID
Explore the full metric catalog

FITS YOUR EXPERIMENT

Your data.
Your metrics.
Your workflow.

01

Bring your audio

Work with file paths, SCP lists, or Kaldi-style ARKs.

02

Configure your evaluation

Select metrics with YAML configurations and install the backends your experiment needs.

03

Run, scale, and explore

Start locally, scale with Slurm, and explore your results with VERSA’s visualization tools.

FROM AUDIO TO INSIGHT

Start your
first evaluation.

Install the toolkit, then try a configuration with the included audio samples. Add optional metric dependencies as needed.

Full installation guide
QUICK START
git clone https://github.com/wavlab-speech/versa.git
cd versa
pip install .

# Run on the included audio samples
python versa/bin/scorer.py \
  --score_config egs/speech_cpu.yaml \
  --pred test/test_samples/test2 \
  --gt test/test_samples/test1 \
  --output_file test_result \
  --io dir
TRY IT

Explore the Colab demo ↗

A hands-on introduction to VERSA.

GO DEEPER

Visualize your results ↗

Explore your evaluation from multiple perspectives.

BUILD ON VERSA

Research & citation ↗

Read and cite the work behind the toolkit.