ਆਵਾਜ਼ ਰਾਹ

Captioned client walkthrough

Hear the sample.
See the evidence boundary.

This short video shows the deployed voice-first interface and the transparent Punjabi test lab. It contains one licensed real-human FLEURS sample. A default selectable caption track and a full text alternative are provided.

Awaaz Raah pilot walkthrough

The video has no narration. Its complete visual description and spoken-audio transcript follow immediately below.

Video transcript and visual description
  1. 0:00–0:05. The deployed home screen shows “Technical evaluation only — live voice disabled.”
  2. 0:05–0:10. The Ontario scope remains visible with verified static human-help links below.
  3. 0:10–0:12. The quality lab appears. It labels the clips as real human read speech and warns that five public clips are a demonstration, not certification.
  4. 0:12–0:16. Real FLEURS Punjabi audio: “ਮੇਰੇ ਲਈ ਇਸਦਾ ਕੋਈ ਮਤਲਬ ਨਹੀਂ ਸੀ; ਇਹ ਸਹੀ ਨਹੀਂ ਸੀ।”
  5. 0:16–0:22. The lab is described: it places the model transcript beside the human reference and reports WER, CER, latency, request ID, and estimated cost.
  6. 0:22–0:28. Evidence boundary: the configured project returned a quota or billing exhaustion response. No model response or quality score is presented.
  7. 0:28–0:36. Public-launch requirement: independent native-speaker testing across Gurmukhi, Shahmukhi, dialects, code-switching, noise, and crisis routing.

Audio: FLEURS corpus by Conneau et al. and Google Research, google/fleurs Punjabi India validation row 3, real human read speech. Distributed under CC BY 4.0; see the dataset card. Five clips were selected, extracted, and renamed without editorial change. It was transcoded and resampled for this MP4; the spoken content is unchanged.