Downloads · 30 days
7
7% of all-time downloads
cdli/whisper-large-v3_finetuned_ghanian_ga_standard_speech_v1.0
whisper-large-v3_finetuned_ghanian_ga_standard_speech_v1.0 is a automatic speech recognition model from cdli. Use it when you need speech turned into text. It is set up for transformers.
This is a fine-tuned version of oopenai/whisper-large-v3 for Ga standard speech. It is part of CDLI's effort to make speech technology work for people whose speech is underserved by mainstream ASR systems.
Downloads · 30 days
7
7% of all-time downloads
All-time downloads
98
Public
Parameters
1.5B
6.2 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors6.2 GB · 100%
From the Hugging Face model README
This is a fine-tuned version of oopenai/whisper-large-v3 for Ga standard speech. It is part of CDLI's effort to make speech technology work for people whose speech is underserved by mainstream ASR systems.
All CDLI models and datasets can be found on CDLI's HuggingFace page.
The model has been fine-tuned using cdli/ghanian_ga_standard_speech_v1.0.
The train split was used for training, and the dev split for selecting the best checkpoint.
This Whisper model was fine-tuned and is decoded using the Yoruba (yo) language setting — out of all languages Whisper supports, the one most similar to Ga.
All model parameters (encoder, decoder, and output projection) were fine-tuned, with SpecAugment enabled.
This model was evaluated on the test split of the dataset. Utterances longer than 30 seconds were excluded:
For decoding we ran Whisper with language=yo, task=transcribe,
greedy search (num_beams=1, do_sample=False).
We report two complementary word error rate (WER) metrics, both computed on text
normalized with Whisper's BasicTextNormalizer:
The per-utterance averaged WER bounds each utterance to [0, 1] and weights all utterances equally, so it reflects typical performance without a few catastrophic utterances dominating — but it is not a true error rate and isn't directly comparable to other published WER, hence we report the standard, corpus-level WER as well.
| Model | Standard WER | Per-utterance averaged WER |
|---|---|---|
| This model | 0.16 | 0.16 |
Aggregated results can hide important underlying patterns, so we also break the WER down by subset: per speaker, and — where speaker severity is available — per impairment severity group.
Per-utterance averaged WER per speaker. n_utterances is the number of test utterances for that speaker.
| speaker_id | n_utterances | Avg WER |
|---|---|---|
| 496 | 107 | 0.12 |
| 520 | 82 | 0.2 |
| 531 | 95 | 0.14 |
| 533 | 115 | 0.16 |
| 538 | 66 | 0.12 |
| 541 | 133 | 0.21 |
| 545 | 47 | 0.15 |
| 546 | 93 | 0.17 |
| 562 | 107 | 0.14 |
| 579 | 114 | 0.24 |
| 587 | 86 | 0.13 |
| 602 | 115 | 0.12 |
| 622 | 40 | 0.11 |
| 625 | 63 | 0.11 |
| 644 | 133 | 0.16 |
| 651 | 64 | 0.11 |
| 666 | 73 | 0.19 |
| 675 | 105 | 0.2 |
| 693 | 106 | 0.24 |
| 707 | 124 | 0.07 |
| 736 | 111 | 0.15 |
| 780 | 46 | 0.2 |