windanam_mms-1b-tts_v2
This model is a fine-tuned version of facebook/mms-1b-all on FulaSpeechCorpora-splited-noi_augmented and FulaNewsTextCorporaTTS dataset. It achieves the following results on the evaluation set:
- Loss: 0.4020
- Wer: 0.2416
If you want to test quickly the model, instead of using the huggingface Inference API
, use this space: https://huggingface.co/spaces/cawoylel/MMS-ASR-Fula
How to use
Load Model
pipe = pipeline("automatic-speech-recognition", model="cawoylel/windanam_mms-1b-tts_v2")
Run Model
transcription = pipe(your_audio_file)
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 100000
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
0.8993 | 0.06 | 2500 | 0.6557 | 0.3266 |
0.8056 | 0.12 | 5000 | 0.5806 | 0.3130 |
0.7663 | 0.17 | 7500 | 0.5514 | 0.2804 |
0.736 | 0.23 | 10000 | 0.5333 | 0.2729 |
0.7111 | 0.29 | 12500 | 0.5200 | 0.2686 |
0.6867 | 0.35 | 15000 | 0.5043 | 0.2657 |
0.6942 | 0.41 | 17500 | 0.4933 | 0.2617 |
0.6695 | 0.47 | 20000 | 0.4814 | 0.2595 |
0.6464 | 0.52 | 22500 | 0.4733 | 0.2576 |
0.694 | 0.58 | 25000 | 0.4618 | 0.2553 |
0.6223 | 0.64 | 27500 | 0.4574 | 0.2544 |
0.6692 | 0.7 | 30000 | 0.4495 | 0.2526 |
0.6337 | 0.76 | 32500 | 0.4454 | 0.2518 |
0.6055 | 0.81 | 35000 | 0.4396 | 0.2501 |
0.6436 | 0.87 | 37500 | 0.4341 | 0.2498 |
0.6389 | 0.93 | 40000 | 0.4304 | 0.2488 |
0.6308 | 0.99 | 42500 | 0.4260 | 0.2486 |
0.6167 | 1.05 | 45000 | 0.4248 | 0.2469 |
0.6253 | 1.11 | 47500 | 0.4201 | 0.2467 |
0.603 | 1.16 | 50000 | 0.4179 | 0.2457 |
0.5864 | 1.22 | 52500 | 0.4187 | 0.2449 |
0.6276 | 1.28 | 55000 | 0.4147 | 0.2446 |
0.5901 | 1.34 | 57500 | 0.4151 | 0.2439 |
0.5877 | 1.4 | 60000 | 0.4126 | 0.2436 |
0.604 | 1.45 | 62500 | 0.4111 | 0.2435 |
0.6182 | 1.51 | 65000 | 0.4097 | 0.2436 |
0.5745 | 1.57 | 67500 | 0.4074 | 0.2433 |
0.5843 | 1.63 | 70000 | 0.4076 | 0.2429 |
0.578 | 1.69 | 72500 | 0.4062 | 0.2429 |
0.5799 | 1.75 | 75000 | 0.4060 | 0.2425 |
0.6017 | 1.8 | 77500 | 0.4047 | 0.2423 |
0.5688 | 1.86 | 80000 | 0.4052 | 0.2417 |
0.5837 | 1.92 | 82500 | 0.4039 | 0.2418 |
0.5801 | 1.98 | 85000 | 0.4038 | 0.2414 |
0.5965 | 2.04 | 87500 | 0.4030 | 0.2413 |
0.5827 | 2.09 | 90000 | 0.4022 | 0.2414 |
0.6043 | 2.15 | 92500 | 0.4020 | 0.2414 |
0.609 | 2.21 | 95000 | 0.4021 | 0.2414 |
0.5827 | 2.27 | 97500 | 0.4021 | 0.2415 |
0.5755 | 2.33 | 100000 | 0.4020 | 0.2416 |
Framework versions
- Transformers 4.36.0.dev0
- Pytorch 2.1.1+cu121
- Datasets 2.15.0
- Tokenizers 0.15.0
- Downloads last month
- 0
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Model tree for cawoylel/windanam_mms-1b-tts_v2
Base model
facebook/mms-1b-all