# Transcribe noisy Hindi-English calls into Roman Hinglish

Use PyAI Hear to transcribe an authorized 8 kHz WAV directly from your backend. The file goes to PyAI over HTTPS without needing a public audio URL. If audio must stay on the device, stop before submitting it.

## Before uploading

Check `GET /v1/me` for the expected project, environment, credits, usage limits and the `hear:transcribe` and `transcribe:jobs` permissions. Keep the API key on your backend and confirm the budget. Async jobs may retain uploaded audio for up to seven days and results for up to 30 days. They do not offer a zero-retention switch. Cancelling a job does not erase its data. [Async job contract](https://docs.pyai.com/guides/async-transcription-jobs) · [Hear pricing](https://pyai.com/pricing)

## Upload the private recording

Use a WAV header matching the original sample rate and channel count. For mixed Hindi and English, leave `transcription_language` unset. For Hindi-only speech, set it to `hi`. The example requests Roman letters with `output_script=latin` and speaker diarization on a mono call. For separate participants on stereo tracks, replace `diarize=true` with `channel=true`.

Set `TRANSCRIPTION_REQUEST_ID` to a unique identifier for this recording and these options. Submit once after authorization.

```sh
curl --fail-with-body -sS https://api.pyai.com/v1/transcription/jobs \
  -H "Authorization: Bearer $PYAI_API_KEY" \
  -H "Idempotency-Key: $TRANSCRIPTION_REQUEST_ID" \
  -F audio=@call-hi-8khz.wav \
  -F output_script=latin \
  -F diarize=true \
  -F output_formats=json
```

## Read the transcript

Save the returned `job_id`. Use the same credential to poll `GET /v1/transcription/jobs/{job_id}` with a time limit. Stop on failure or cancellation. If submission was uncertain, reconcile the original job using its idempotency key before attempting another submission.

On completion read `result.text`, `result.words` and `result.segments`, or download the signed `result_url` when the output is offloaded. Word and segment times are in seconds. Check each segment for a `speaker` value before promising speaker labels. A completed job can return text without them.

## Recorded results

On 21 October 2026, Hear transcribed a 28.4-second synthetic Hindi-English recording. The mono WAV was sampled at 24 kHz. Added background noise matched the level of active speech (0 dB signal-to-noise ratio). Hear preserved the quantity of 24 and changed the invoice amount from 1,499 to 2,400. For amounts, names and reference numbers, use the word timestamps to locate the passage and check it against the recording.

On 17 November 2026 a separate 20.18-second, 16 kHz stereo job with Latin output returned 62 timed words, 23 segments and channel IDs 0 and 1. Channel IDs identify the two audio tracks. Speaker-label accuracy on a noisy 8 kHz mono call remains unmeasured here. Evaluate the speaker changes and code-switched words on a recording representative of your callers.

[Read the original and returned transcript excerpts](https://pyai.com/guides/hindi-call-transcription#transcription-results).
