michsethowusu's picture
Rename README(5).md to README.md
03f30e9 verified

A newer version of the Gradio SDK is available: 6.12.0

Upgrade
metadata
title: ASR Benchmark  Ghanaian Languages
emoji: 🎙️
colorFrom: blue
colorTo: blue
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: apache-2.0
hardware: a10g-small

ASR Benchmark — Ghanaian Languages

Benchmarks Wav2Vec2-BERT fine-tuned models and Gemma-3n on Dagbani, Twi, and Ewe speech datasets.

Models evaluated

Label HF Hub ID
Dagbani (w2v-bert-2.0) FarmerlineML/w2v-bert-2.0_2026_dagbani_ASR
Ewe (w2v-bert-2.0) FarmerlineML/w2v-bert-2.0_ewe_2
Twi (w2v-bert-2.0) FarmerlineML/w2v-bert-2.0_twi_alpha_v1
Gemma-3n (E2B-it) unsloth/gemma-3n-E2B-it

Datasets

Label HF Hub path
UNICEF Dagbani ghananlpcommunity/UNICEF-Ghana-Dagbani-ASR
UNICEF Twi ghananlpcommunity/UNICEF-Ghana-Twi-ASR
UNICEF Ewe ghananlpcommunity/UNICEF-Ghana-Ewe-ASR
Bible Ewe ghananlpcommunity/ewe-bible-audio-text-tts
Bible Dagbani ghananlpcommunity/dagbani-bible-audio-text-tts
Bible Twi ghananlpcommunity/asante-twi-bible-speech-text

Metrics

  • CER — Character Error Rate
  • WER — Word Error Rate

Hardware recommendation

Run on an a10g-small (24 GB VRAM) Space for Gemma-3n 4-bit inference.
Wav2Vec2 models alone can run on cpu-basic or t4-small.