SpeechT5 fine-tuned on MINDS-14 en-US
This model is a fine-tuned version of
microsoft/speecht5_tts for the Hugging Face Audio Course
Unit 6 hands-on assignment.
Training data
Dataset: PolyAI/minds14, English (en-US).
A small subset was used for fine-tuning on a free Google Colab Tesla T4 GPU.
Speaker embeddings were extracted with
speechbrain/spkrec-xvect-voxceleb.
Training
- Base model:
microsoft/speecht5_tts - Task:
text-to-speech - Training steps: 100
- Learning rate: 1e-5
- FP16 training
- Gradient checkpointing enabled during training
Evaluation
Final validation loss: 1.000769
The Hugging Face Audio Course Unit 6 assignment does not specify a minimum evaluation metric. Its objective is to practice SpeechT5 fine-tuning and publish the resulting TTS model to the Hub.
Audio Course
Hugging Face Audio Course — Unit 6: From Text to Speech.
- Downloads last month
- 11
Model tree for herurg/speecht5-minds14-en-us-audio-course
Base model
microsoft/speecht5_tts