This model is a fine-tuned version of openai/whisper-medium on the Common Voice 11.0 dataset, and the Romanian speech synthesis corpus. It achieves the following results on the evaluation set:
The architecture is the same as openai/whisper-medium.
The model was trained on the Common Voice 11.0 dataset (train+validation+other splits) and the Romanian speech synthesis corpus, and was tested on the test split of the Common Voice 11.0 dataset.
Inference with 🤗 transformers
from transformers import WhisperProcessor, WhisperForConditionalGeneration
from datasets import Audio, load_dataset
import torch
# load model and processor
processor = WhisperProcessor.from_pretrained("gigant/whisper-medium-romanian")
model = WhisperForConditionalGeneration.from_pretrained("gigant/whisper-medium-romanian")
# load dummy dataset and read soundfiles
ds = load_dataset("common_voice", "ro", split="test", streaming=True)
ds = ds.cast_column("audio", Audio(sampling_rate=16_000))
input_speech = next(iter(ds))["audio"]["array"]
model.config.forced_decoder_ids = processor.get_decoder_prompt_ids(language = "ro", task = "transcribe")
input_features = processor(input_speech, return_tensors="pt", sampling_rate=16_000).input_features
predicted_ids = model.generate(input_features, max_length=448)
# transcription = processor.batch_decode(predicted_ids)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens = True)
The code was adapted from openai/whisper-medium.
The following hyperparameters were used during training:
checkpoint-1000/optimizer.pt 6.1 GBcheckpoint-1500/optimizer.pt 6.1 GBcheckpoint-2000/optimizer.pt 6.1 GBcheckpoint-2500/optimizer.pt 6.1 GBcheckpoint-3000/optimizer.pt 6.1 GBcheckpoint-3500/optimizer.pt 6.1 GBcheckpoint-500/optimizer.pt 6.1 GBcheckpoint-1000/pytorch_model.bin 3.1 GBcheckpoint-1500/pytorch_model.bin 3.1 GBcheckpoint-2000/pytorch_model.bin 3.1 GBcheckpoint-2500/pytorch_model.bin 3.1 GBcheckpoint-3000/pytorch_model.bin 3.1 GBcheckpoint-3500/pytorch_model.bin 3.1 GBcheckpoint-500/pytorch_model.bin 3.1 GBpytorch_model.bin 3.1 GBmodel.safetensors 3.1 GBcheckpoint-3500/rng_state.pth 14.6 kBcheckpoint-1000/rng_state.pth 14.6 kBcheckpoint-1500/rng_state.pth 14.6 kBcheckpoint-2000/rng_state.pth 14.6 kBcheckpoint-2500/rng_state.pth 14.6 kBcheckpoint-3000/rng_state.pth 14.6 kBcheckpoint-500/rng_state.pth 14.6 kBcheckpoint-1000/training_args.bin 3.5 kBcheckpoint-1500/training_args.bin 3.5 kBcheckpoint-2000/training_args.bin 3.5 kBcheckpoint-2500/training_args.bin 3.5 kBcheckpoint-3000/training_args.bin 3.5 kBcheckpoint-3500/training_args.bin 3.5 kBcheckpoint-500/training_args.bin 3.5 kBtraining_args.bin 3.5 kBcheckpoint-1000/scheduler.pt 0.6 kBcheckpoint-1500/scheduler.pt 0.6 kBcheckpoint-2000/scheduler.pt 0.6 kBcheckpoint-2500/scheduler.pt 0.6 kBcheckpoint-3000/scheduler.pt 0.6 kBcheckpoint-3500/scheduler.pt 0.6 kBcheckpoint-500/scheduler.pt 0.6 kBcheckpoint-1000/scaler.pt 0.6 kBcheckpoint-1500/scaler.pt 0.6 kBcheckpoint-2000/scaler.pt 0.6 kBcheckpoint-2500/scaler.pt 0.6 kBcheckpoint-3000/scaler.pt 0.6 kBcheckpoint-3500/scaler.pt 0.6 kBcheckpoint-500/scaler.pt 0.6 kB有文件?提交种子。
提交种子openai/whisper-medium