PrinceAlhassanNasamu/whisper-large-v3-turbo-tekyerema-eng

Ghanaian English ASR (Whisper large-v3-turbo)

Author: Prince Nasamu Alhassan

Overview

Fine-tuned from openai/whisper-large-v3-turbo on Ghanaian English. The character error rate is less than half the baseline's, which matters more than the word rate for a system that has to recover a phone number or an amount from what it heard.

This is not the best checkpoint this run produced. Evaluated every 1,000 steps, it scored 16.53 WER at step 1000, 17.59 at 2000 and 17.65 at 3000. The step-1000 checkpoint was deleted by save_total_limit=1 before anyone thought to preserve it, and load_best_model_at_end was not set on this run -- it was added to the training script afterwards, too late for this model. What is published here is the step-2000 checkpoint, one point worse than a model that no longer exists. The 16.53 is recorded because it was observed, not because it is available.

The run was then stopped with 7,000 steps left: three evaluations showed it had peaked and flattened, and the GPU was needed elsewhere.

Use it

import torch, soundfile as sf
from transformers import AutoProcessor, WhisperForConditionalGeneration

repo  = "PrinceAlhassanNasamu/whisper-large-v3-turbo-tekyerema-eng"
proc  = AutoProcessor.from_pretrained(repo)
# fp32 on purpose. The checkpoint is stored fp16 and audio features are
# fp32, so the first conv otherwise fails with "Input type (float) and bias
# type (c10::Half) should be the same" — and CPUs cannot do fp16 conv
# usefully anyway.
model = WhisperForConditionalGeneration.from_pretrained(
    repo, dtype=torch.float32).eval()

wav, sr = sf.read("clip.wav", dtype="float32")   # 16 kHz mono
feats = proc(wav, sampling_rate=16_000, return_tensors="pt").input_features
with torch.no_grad():
    ids = model.generate(feats, language="en", task="transcribe")
print(proc.batch_decode(ids, skip_special_tokens=True)[0].strip())

Training data

Trained on the Ghana Speech dataset and related Ghanaian corpora, licensed CC BY-NC 4.0.

Measured

On afrispeech_ghana, same items and same scorer as the baseline:

model	WER / CER
baseline it was fine-tuned from	29.74 / 16.71
this model	17.85 / 7.29

A WER alone is not informative — compared against another language it means nothing. Compared against the model it started from, it means everything.

Intended use & license

Non-commercial use only (CC BY-NC 4.0). This is inherited from the training data and required by the terms under which the compute was granted: models trained in that window are non-commercial by condition of access, not by inference.

Limitations, stated plainly

• Dagbani had no recogniser of its own for this whole project, and the reason given for that was wrong. Every card here said "one fine-tuning session on 74 validation rows would not change that". Those 74 rows are the eng-dag machine-translation validation split. The Dagbani speech data in this same account is waxal_dag: 13,228 training rows, 1,750 validation rows, ~71 hours, 1,041 speakers with the largest at 1% — more data and better speaker diversity than Ewe, which produced a working 42.19 WER recogniser. A number was carried across from a translation table into a speech claim, and then repeated on every model card on the account. It is training now, on 2026-08-31. Until it is scored, the honest statement is that Dagbani's best available recogniser scores 86.6 WER and nobody had tried fine-tuning on the data already in hand.
• Evaluation is on read and machine-translated text. No recordings of people speaking agent commands in these languages exist. Numbers measured this way are optimistic about phrasing and pessimistic about code-switching, and should not be read as field performance.
• Research work from a hackathon entry, not a supported product.

The rest of the family

Recognisers

• whisper-large-v3-turbo-tekyerema-eng-foundation — Ghanaian English ASR — course 1 (foundation)
• kusaal-whisper-small-lora — Kusaal ASR (Whisper-small LoRA, superseded)
• kasa42-asr — KASA-42 (Kusaal, third-party export)
• tekyerema-asr-ctc — Twi ASR (w2v-BERT CTC)
• tekyerema-asr-mms-ewe — Ewe ASR (MMS adapter)
• tekyerema-asr-mms-dag — Dagbani ASR (MMS adapter)
• tekyerema-asr-mms-hau — Hausa ASR (MMS adapter)
• tekyerema-asr-mms-kus — Kusaal ASR (MMS adapter)
• whisper-large-v3-turbo-tekyerema-eng — Ghanaian English ASR (Whisper large-v3-turbo)

Voices

• tekyerema-tts-twi — Twi TTS (VITS)
• tekyerema-tts-kus — Kusaal TTS (VITS)
• tekyerema-tts-ewe — Ewe TTS (VITS)
• tekyerema-tts-hau — Hausa TTS (VITS)
• tekyerema-tts-eng — Ghanaian English TTS (VITS)

Agent models

• tekyerema-1-reply — Tɛkyerɛma-1 reply adapter (arm ①)
• tekyerema-1-native-reply — Tɛkyerɛma-1 reply adapter (arm ②)
• tekyerema-1-tool — Tɛkyerɛma-1 tool adapter (arm 1)
• tekyerema-audio-native — Tɛkyerɛma-1 audio-native (arm 3)
• tekyerema-audio-native-4k — Tɛkyerɛma-1 audio-native, 4,000 clips (arm 3 v2)
• tekyerema-1-native-tool — Tɛkyerɛma-1 tool adapter (arm 2)

Translation

• tekyerema-nllb600m-v1 — Tɛkyerɛma MT v1 (NLLB-600M)
• kusaal-nllb-600M — Kusaal MT specialist (NLLB-600M)

Routing

• tekyerema-intent-afroxlmr — Intent classifier (AfroXLMR)

Acknowledgements

Compute resources provided by AI Skills and Compute Africa (AISCA). Trained on the Ghana NLP H200 GPU. Please keep derivatives non-commercial and share improvements back with the Ghana NLP community (ghananlpcommunity).