umarigan/speecht5_tts_tr_v1.0

SpeechT5 TTS Turkish

This model is a fine-tuned version of microsoft/speecht5_tts on the turkishvoicedataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.3079

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.4436	1.8484	1000	0.3752
0.3822	3.6969	2000	0.3403
0.3729	5.5453	3000	0.3233
0.3451	7.3937	4000	0.3153
0.3315	9.2421	5000	0.3099
0.3492	11.0906	6000	0.3079

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1

Usage

installs

!pip install datasets soundfile speechbrain

inference

from transformers import pipeline
from datasets import load_dataset
import soundfile as sf
import torch
from IPython.display import Audio

synthesiser = pipeline("text-to-speech", "umarigan/speecht5_tts_tr_v1.0")

embeddings_dataset = load_dataset("umarigan/turkish_voice_dataset_embedded", split="train")
speaker_embedding = torch.tensor(embeddings_dataset[736]["speaker_embeddings"]).unsqueeze(0)

# Synthesize speech using the embedding
speech = synthesiser("Bir berber bir berbere gel beraber bir berber kuralım demiş", forward_params={"speaker_embeddings": speaker_embedding})

# Save the generated audio to a file
sf.write("speech.wav", speech["audio"], samplerate=speech["sampling_rate"])

# Play the audio in the notebook
Audio("speech.wav")