---
title: Hausa Voice Lab
emoji: 🎙️
colorFrom: green
colorTo: blue
sdk: gradio
sdk_version: 6.27.0
app_file: app.py
python_version: "3.12"
short_description: Record Hausa speech and transcribe it with BuzzASR
startup_duration_timeout: 30m
license: mit
models:
  - BuzzASR/hausa
tags:
  - hausa
  - speech-recognition
  - audio
  - zerogpu
---

# Hausa Voice Lab

A microphone and audio-upload demo for [BuzzASR/hausa](https://huggingface.co/BuzzASR/hausa),
a Hausa-specialized Whisper Large v3 model released by lemn-lab under the MIT license.

Record or upload **up to 30 seconds** of Hausa speech, press **Transcribe Hausa**,
then copy the result or download a UTF-8 text file. The app preserves the model's
output: it does not replace names, translate, identify speakers, or use Gemini.

The model is Hausa-only. English code-switching, names, noise, and dialects can
reduce accuracy. Compare with your recording rather than treating the transcript
as verified fact. No artificial example audio is presented as a real evaluation.

## Hosting

Deploy as a Gradio Space on **ZeroGPU** (`zero-a10g`). Model weights load once at
startup. The transcription handler requests GPU time only for inference. The
initial 60-second GPU reservation should be tuned after measuring live calls.
Free hosting depends on account eligibility and available ZeroGPU slots; visitors
are subject to Hugging Face queues and GPU quotas. This is not unlimited hosting.

Audio is sent to this Space for processing. Gradio temporarily caches uploads and
generated transcript files; this app schedules cache cleanup every five minutes
for files older than one hour. It does not add recordings to datasets, application
logs, or permanent storage. Do not upload recordings you do not have permission to share.

## Deployment

From the main project directory, after `hf auth login`:

```powershell
.\.hf-tools\Scripts\hf.exe repos create YOUR_USERNAME/buzzasr-hausa-demo --type space --space-sdk gradio --flavor zero-a10g --public
.\.hf-tools\Scripts\hf.exe upload YOUR_USERNAME/buzzasr-hausa-demo spaces/buzzasr-hausa/app.py app.py --repo-type space
.\.hf-tools\Scripts\hf.exe upload YOUR_USERNAME/buzzasr-hausa-demo spaces/buzzasr-hausa/requirements.txt requirements.txt --repo-type space
.\.hf-tools\Scripts\hf.exe upload YOUR_USERNAME/buzzasr-hausa-demo spaces/buzzasr-hausa/README.md README.md --repo-type space
```

Only upload this folder, never the parent app or its `.env.local`.

The app exposes `/transcribe` through Gradio's API and enables its MCP server.
Deployment verification must include a real audio request and inspection of the
transcript and runtime logs, not only a successful build.