aktasumitt / Translate_Audio_English_To_Turkish

We Use Concatinate three models for translate speech from English to Turkish

Translate Audio from English to Turkish

Introduction:

• In this project, I aimed to translate English Audio to Turkish Audio.
• In this project, I used separate models for English Speech to Text, Text Translation to Turkish, and Turkish Text to Speech, and integrated them all into one place. When run sequentially, we achieved audio-to-audio translation.
• For Speech to Text, I utilized the OpenAI Whisper model, the Bart Model for translation, and the Bark model for Text to Speech.

For details :

• Whisper : https://openai.com/research/whisper
• Bark: https://huggingface.co/suno/bark
• Bart: https://huggingface.co/docs/transformers/model_doc/bart

Dataset:

• I didnt train the models. I just use for prediction so we have just audio for predict.

Train:

• firstly, we use pretrained Whisper Model for English speech to text
• Secondly, we use pretrained Bart Model for translating from English text that is output of Whisper to Turkish
• After that , we use pretrained Bark Model with the output of Bart for transforming from text to speech.
• I use New York City Introduction audio to transform

Usage:

• You can use directly Audio to Audio with Concatinated_Models folder. You may need to set paths in config file acording to your Data paths
• If you want to use the models separately, you can go to their respective folders and use only what you need from there
• Predictions of Models will save "Prediction" folder