speecht5_tr_commonvoice_2

This model is a fine-tuned version of microsoft/speecht5_tts on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5934

Model description

import torch
from transformers import pipeline
from datasets import load_dataset
import soundfile as sf

embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embedding = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)

from transformers import pipeline
pipe = pipeline("text-to-audio", model="Chan-Y/speecht5_finetuned_tr_commonvoice")

text = "bugün okula erken geldim, çalışmam lazım. çok sıkıcı bir dersim var."
result = pipe(text, forward_params={"speaker_embeddings": speaker_embedding})

sf.write("speech.wav", result["audio"], samplerate=result["sampling_rate"])

from IPython.display import Audio
Audio("speech.wav")

Training and evaluation data

I used CommonVoice Turkish Corpus 19.0

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-06
• train_batch_size: 8
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.7533	1.2972	1000	0.6445
0.6745	2.5945	2000	0.6106
0.6535	3.8917	3000	0.5953
0.6593	5.1889	4000	0.5934

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3