teckedd/gha-dondo-w2v-bert-twi-v2

DONDO w2v-BERT CTC fine-tune trial for Ghana Health AI (dondo-twi-v2). Promotion candidate: True.

• Base model: KhayaAI/w2v-bert-ada_ewe_fat_fra_gaa_nzi_twi_en
• Task: automatic-speech-recognition
• Languages: tw, ak
• Card generated: 2026-08-18 (UTC)
• Product: ghanahealth.serendepify.com

Not a medical device. Outputs support community health guidance only.

Metrics

Metric	Value
val_wer	0.2743
val_cer	0.0902

Training data

• google/WaxalNLP (aka_asr)
• fsicoli/common_voice_22_0 (tw)
• local (ghana-health-ai-recorder)

Intended use

• In-product ASR / TTS / chat for Twi (Akan) and English health conversations in Ghana.
• Research on low-resource Ghanaian language speech and health dialogue.

Out of scope

• Clinical diagnosis or autonomous medical decisions.
• Claiming near-native quality without reporting held-out WER/CER or human A/B scores.

How to load

from transformers import pipeline  # or AutoModel + processor per task
# repo: teckedd/gha-dondo-w2v-bert-twi-v2

Citation

If you use this checkpoint, please credit Ghana Health AI / Serendepify and the upstream base model authors plus any listed datasets.

DONDO language prefix

This checkpoint follows DONDO's language-conditioned CTC setup. For Asante Twi, prepend language id 2 to acoustic features before decoding.

Training mix

• Sources: google/WaxalNLP:aka_asr(n=10107), fsicoli/common_voice_22_0:tw(n=201), local:ghana-health-ai-recorder(n=32)
• Learning rate: 5e-05 · max steps: 2500

Decode

• Greedy CTC: WER 0.27425920666477327 / CER 0.09021929127602098
• Greedy CTC argmax only (no LM decode in this run).

Promotion gate

• Current v6 Whisper beam=5 WER: 0.3044
• This validation WER (greedy): 0.27425920666477327
• Promote: True