Automatic Speech Recognition
Transformers
Safetensors
Amharic
wav2vec2-bert
amharic
ethiopic
telecom
ctc
kenlm
pyctcdecode
ethio-asr
Instructions to use SaarAI/Amharic-ASR-v2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use SaarAI/Amharic-ASR-v2 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("automatic-speech-recognition", model="SaarAI/Amharic-ASR-v2")# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForCTC processor = AutoProcessor.from_pretrained("SaarAI/Amharic-ASR-v2") model = AutoModelForCTC.from_pretrained("SaarAI/Amharic-ASR-v2", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Model card: public release
Browse files
README.md
CHANGED
|
@@ -36,7 +36,7 @@ tags:
|
|
| 36 |
|
| 37 |
# Amharic-ASR-v2
|
| 38 |
|
| 39 |
-
>
|
| 40 |
> (9 test sets). v2 + LM has the **best average WER (27.15%)** of all models
|
| 41 |
> tested, ahead of v1 (28.19%) and `addisai/addis-scribe-streaming` (31.62%).
|
| 42 |
|
|
@@ -77,7 +77,7 @@ pip install transformers torch torchaudio pyctcdecode kenlm # kenlm needs Pyth
|
|
| 77 |
from transformers import pipeline
|
| 78 |
|
| 79 |
asr = pipeline("automatic-speech-recognition",
|
| 80 |
-
model="SaarAI/Amharic-ASR-v2",
|
| 81 |
device=0)
|
| 82 |
print(asr("audio.wav")["text"]) # uses the LM
|
| 83 |
print(asr("call.wav", chunk_length_s=20, stride_length_s=4)["text"]) # long audio
|
|
|
|
| 36 |
|
| 37 |
# Amharic-ASR-v2
|
| 38 |
|
| 39 |
+
> Benchmarked on the full SaarAI Amharic ASR leaderboard
|
| 40 |
> (9 test sets). v2 + LM has the **best average WER (27.15%)** of all models
|
| 41 |
> tested, ahead of v1 (28.19%) and `addisai/addis-scribe-streaming` (31.62%).
|
| 42 |
|
|
|
|
| 77 |
from transformers import pipeline
|
| 78 |
|
| 79 |
asr = pipeline("automatic-speech-recognition",
|
| 80 |
+
model="SaarAI/Amharic-ASR-v2",
|
| 81 |
device=0)
|
| 82 |
print(asr("audio.wav")["text"]) # uses the LM
|
| 83 |
print(asr("call.wav", chunk_length_s=20, stride_length_s=4)["text"]) # long audio
|