---
license: mit
base_model: microsoft/speecht5_tts
tags:
- generated_from_trainer
- turkish
- tr
model-index:
- name: speecht5_finetuned_emirhan_tr
  results: []
language:
- tr
pipeline_tag: text-to-speech
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# speecht5_finetuned_emirhan_tr

This model is a fine-tuned version of [microsoft/speecht5_tts](https://huggingface.co/microsoft/speecht5_tts) on [erenfazlioglu/turkishvoicedataset](https://huggingface.co/datasets/erenfazlioglu/turkishvoicedataset).
It achieves the following results on the evaluation set:
- Loss: 0.3135

## Model description

The base model uses a transformer-based approach, specifically Transfer Transformer, to generate high-quality speech from text. The fine-tuning process on the Turkish Voice Dataset enables the model to produce more natural and accurate speech in Turkish.

## Intended uses & limitations

This model is intended for text-to-speech (TTS) applications specifically tailored for the Turkish language. It can be used in various scenarios, such as voice assistants, automated announcements, and accessibility tools for Turkish speakers.

## Training and evaluation data

The model's performance is optimized for Turkish and may not generalize well to other languages.
The model might not handle rare or domain-specific vocabulary as effectively as more common words.

## Training procedure

The model was fine-tuned on the Turkish Voice Dataset, which consists of high-quality synthetic Turkish voice recordings from Microsoft Azure. The dataset was split into training and evaluation subsets, with the evaluation set used to measure the model's loss and overall performance.


### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 4
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 8
- 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: 100
- training_steps: 660
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch  | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 0.514         | 0.4545 | 100  | 0.4372          |
| 0.4226        | 0.9091 | 200  | 0.3626          |
| 0.3771        | 1.3636 | 300  | 0.3417          |
| 0.3562        | 1.8182 | 400  | 0.3278          |
| 0.3472        | 2.2727 | 500  | 0.3217          |
| 0.3402        | 2.7273 | 600  | 0.3135          |


### Framework versions

- Transformers 4.42.4
- Pytorch 2.4.0+cu121
- Datasets 2.21.0
- Tokenizers 0.19.1