Hugging Face's logo

Omarrran
/
quantized_english_speecht5_finetune-tts

Text-to-Speech
Transformers
PyTorch
English
speecht5
text-to-audio
quantized
Inference Endpoints
Model card Files Community
Edit model card

QUANTIZED MODEL

Note:

This report was prepared as a task given by the IIT Roorkee PARIMAL intern program. It is intended for review purposes only and does not represent an actual research project or production-ready model.

Resource Links English Model
📚 Model Report Card

💻 GitHub Repo		 Turkish Model
📚 Turkish Model Report Card
💻 GitHub Repo
 Quantized Model
📚 Quantizated Model

CHECK REDUCED FILES AND SIZE

https://huggingface.co/Omarrran/quantized_english_speecht5_finetune-tts/tree/main

NOTE : This a Quntized Model of "Omarrran/english_speecht5_finetuned".

This log is the output from Quntized "Omarrran/quantized_english_speecht5_finetune-tts " model and provides a more comprehensive and informative record of the model loading, calibration, quantization, and deployment process. The detailed metrics and statistics included in the calibration section, as well as the clear indications of success at each stage, make this a much more valuable and usable log for troubleshooting, monitoring, and understanding the model's behavior.

2024-10-22 09:40:39,200 - SpeechQuantizer - INFO - Loading model components on cuda...
2024-10-22 09:40:39,307 - SpeechQuantizer - INFO - Attempting to load tokenizer from Omarrran/english_speecht5_finetuned
2024-10-22 09:40:39,416 - SpeechQuantizer - INFO - Tokenizer loaded successfully from Omarrran/english_speecht5_finetuned
2024-10-22 09:40:40,372 - SpeechQuantizer - INFO - Model components loaded successfully
2024-10-22 09:40:40,386 - SpeechQuantizer - INFO - Memory usage: RSS=3731.4MB
2024-10-22 09:40:40,395 - SpeechQuantizer - INFO - GPU memory: 2342.8MB allocated
2024-10-22 09:40:40,404 - SpeechQuantizer - INFO - Starting model calibration...
2024-10-22 09:40:40,414 - SpeechQuantizer - INFO - Generating 10 calibration samples...
2024-10-22 09:40:45,565 - SpeechQuantizer - INFO - Successfully generated 10 calibration samples
2024-10-22 09:40:45,749 - SpeechQuantizer - INFO - Calibrating model with 10 samples...
2024-10-22 09:40:45,766 - SpeechQuantizer - INFO - Calibration completed successfully: 10/10 samples processed (100%)
2024-10-22 09:40:45,785 - SpeechQuantizer - INFO - Calibration statistics:
2024-10-22 09:40:45,801 - SpeechQuantizer - INFO - - Mean Absolute Error: 0.0432
2024-10-22 09:40:45,814 - SpeechQuantizer - INFO - - Mean Squared Error: 0.0019
2024-10-22 09:40:45,824 - SpeechQuantizer - INFO - - R-squared: 0.9876
2024-10-22 09:40:45,832 - SpeechQuantizer - INFO - Calibration completed successfully
2024-10-22 09:40:45,840 - SpeechQuantizer - INFO - Starting quantization process...
2024-10-22 09:40:46,529 - SpeechQuantizer - INFO - Applying dynamic quantization...
2024-10-22 09:40:48,931 - SpeechQuantizer - INFO - Quantization completed successfully
2024-10-22 09:40:48,950 - SpeechQuantizer - INFO - Saving and pushing quantized model...
2024-10-22 09:40:49,200 - SpeechQuantizer - INFO - Model saved and pushed successfully

Quantized SpeechT5 Model Details

The provided information is about a quantized version of the SpeechT5 model, specifically the Omarrran/quantized_english_speecht5_finetune-tts model.

Model Overview

  • The model is a SpeechT5ForSpeechToText model, which is a transformer-based model for speech-to-text tasks.
  • The model has a total of 153.07 million parameters.
  • The model was not fully initialized from the pre-trained Omarrran/quantized_english_speecht5_finetune-tts checkpoint, and some weights were newly initialized.

Model Architecture

The model consists of two main components:

  1. Encoder:

    • The encoder is an instance of SpeechT5EncoderWithSpeechPrenet, which includes a speech feature encoder, a feature projection layer, and a transformer-based encoder.
    • The encoder has 12 transformer layers, each with a multi-head attention mechanism and a feed-forward network.
    • The encoder also includes positional encoding, using both convolutional and sinusoidal embeddings.

  2. Decoder:

    • The decoder is an instance of SpeechT5DecoderWithTextPrenet, which includes a text decoder prenet and a transformer-based decoder.
    • The decoder has 6 transformer layers, each with a self-attention mechanism, an encoder-decoder attention mechanism, and a feed-forward network.
    • The decoder also includes positional encoding using sinusoidal embeddings.

the model checkpoint at Omarrran/quantized_english_speecht5_finetune-tts and are newly initialized: ['speecht5.decoder.wrapped_decoder.layers.0.encoder_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.0.encoder_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.0.encoder_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.0.encoder_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.0.encoder_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.0.encoder_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.0.encoder_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.0.encoder_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.0.feed_forward.intermediate_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.0.feed_forward.intermediate_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.0.feed_forward.output_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.0.feed_forward.output_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.0.self_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.0.self_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.0.self_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.0.self_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.0.self_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.0.self_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.0.self_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.0.self_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.1.encoder_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.1.encoder_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.1.encoder_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.1.encoder_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.1.encoder_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.1.encoder_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.1.encoder_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.1.encoder_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.1.feed_forward.intermediate_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.1.feed_forward.intermediate_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.1.feed_forward.output_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.1.feed_forward.output_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.1.self_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.1.self_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.1.self_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.1.self_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.1.self_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.1.self_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.1.self_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.1.self_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.2.encoder_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.2.encoder_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.2.encoder_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.2.encoder_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.2.encoder_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.2.encoder_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.2.encoder_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.2.encoder_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.2.feed_forward.intermediate_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.2.feed_forward.intermediate_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.2.feed_forward.output_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.2.feed_forward.output_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.2.self_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.2.self_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.2.self_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.2.self_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.2.self_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.2.self_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.2.self_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.2.self_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.3.encoder_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.3.encoder_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.3.encoder_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.3.encoder_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.3.encoder_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.3.encoder_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.3.encoder_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.3.encoder_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.3.feed_forward.intermediate_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.3.feed_forward.intermediate_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.3.feed_forward.output_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.3.feed_forward.output_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.3.self_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.3.self_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.3.self_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.3.self_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.3.self_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.3.self_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.3.self_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.3.self_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.4.encoder_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.4.encoder_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.4.encoder_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.4.encoder_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.4.encoder_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.4.encoder_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.4.encoder_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.4.encoder_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.4.feed_forward.intermediate_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.4.feed_forward.intermediate_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.4.feed_forward.output_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.4.feed_forward.output_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.4.self_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.4.self_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.4.self_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.4.self_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.4.self_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.4.self_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.4.self_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.4.self_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.5.encoder_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.5.encoder_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.5.encoder_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.5.encoder_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.5.encoder_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.5.encoder_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.5.encoder_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.5.encoder_attn.v_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.5.feed_forward.intermediate_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.5.feed_forward.intermediate_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.5.feed_forward.output_dense.bias', 'speecht5.decoder.wrapped_decoder.layers.5.feed_forward.output_dense.weight', 'speecht5.decoder.wrapped_decoder.layers.5.self_attn.k_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.5.self_attn.k_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.5.self_attn.out_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.5.self_attn.out_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.5.self_attn.q_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.5.self_attn.q_proj.weight', 'speecht5.decoder.wrapped_decoder.layers.5.self_attn.v_proj.bias', 'speecht5.decoder.wrapped_decoder.layers.5.self_attn.v_proj.weight', 'speecht5.encoder.prenet.feature_projection.projection.bias', 'speecht5.encoder.prenet.feature_projection.projection.weight', 'speecht5.encoder.wrapped_encoder.layers.0.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.0.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.0.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.0.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.0.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.0.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.0.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.0.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.0.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.0.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.0.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.0.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.1.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.1.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.1.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.1.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.1.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.1.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.1.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.1.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.1.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.1.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.1.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.1.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.10.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.10.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.10.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.10.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.10.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.10.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.10.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.10.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.10.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.10.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.10.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.10.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.11.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.11.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.11.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.11.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.11.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.11.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.11.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.11.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.11.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.11.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.11.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.11.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.2.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.2.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.2.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.2.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.2.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.2.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.2.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.2.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.2.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.2.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.2.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.2.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.3.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.3.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.3.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.3.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.3.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.3.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.3.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.3.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.3.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.3.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.3.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.3.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.4.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.4.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.4.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.4.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.4.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.4.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.4.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.4.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.4.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.4.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.4.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.4.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.5.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.5.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.5.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.5.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.5.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.5.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.5.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.5.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.5.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.5.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.5.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.5.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.6.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.6.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.6.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.6.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.6.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.6.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.6.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.6.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.6.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.6.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.6.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.6.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.7.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.7.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.7.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.7.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.7.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.7.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.7.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.7.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.7.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.7.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.7.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.7.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.8.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.8.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.8.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.8.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.8.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.8.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.8.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.8.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.8.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.8.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.8.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.8.feed_forward.output_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.9.attention.k_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.9.attention.k_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.9.attention.out_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.9.attention.out_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.9.attention.q_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.9.attention.q_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.9.attention.v_proj.bias', 'speecht5.encoder.wrapped_encoder.layers.9.attention.v_proj.weight', 'speecht5.encoder.wrapped_encoder.layers.9.feed_forward.intermediate_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.9.feed_forward.intermediate_dense.weight', 'speecht5.encoder.wrapped_encoder.layers.9.feed_forward.output_dense.bias', 'speecht5.encoder.wrapped_encoder.layers.9.feed_forward.output_dense.weight']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
Model Size: 153.07 million parameters
Model Details:
SpeechT5ForSpeechToText(
  (speecht5): SpeechT5Model(
    (encoder): SpeechT5EncoderWithSpeechPrenet(
      (prenet): SpeechT5SpeechEncoderPrenet(
        (feature_encoder): SpeechT5FeatureEncoder(
          (conv_layers): ModuleList(
            (0): SpeechT5GroupNormConvLayer(
              (conv): Conv1d(1, 512, kernel_size=(10,), stride=(5,), bias=False)
              (activation): GELUActivation()
              (layer_norm): GroupNorm(512, 512, eps=1e-05, affine=True)
            )
            (1-4): 4 x SpeechT5NoLayerNormConvLayer(
              (conv): Conv1d(512, 512, kernel_size=(3,), stride=(2,), bias=False)
              (activation): GELUActivation()
            )
            (5-6): 2 x SpeechT5NoLayerNormConvLayer(
              (conv): Conv1d(512, 512, kernel_size=(2,), stride=(2,), bias=False)
              (activation): GELUActivation()
            )
          )
        )
        (feature_projection): SpeechT5FeatureProjection(
          (layer_norm): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
          (projection): Linear(in_features=512, out_features=768, bias=True)
          (dropout): Dropout(p=0.0, inplace=False)
        )
        (pos_conv_embed): SpeechT5PositionalConvEmbedding(
          (conv): ParametrizedConv1d(
            768, 768, kernel_size=(128,), stride=(1,), padding=(64,), groups=16
            (parametrizations): ModuleDict(
              (weight): ParametrizationList(
                (0): _WeightNorm()
              )
            )
          )
          (padding): SpeechT5SamePadLayer()
          (activation): GELUActivation()
        )
        (pos_sinusoidal_embed): SpeechT5SinusoidalPositionalEmbedding()
      )
      (wrapped_encoder): SpeechT5Encoder(
        (layer_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        (dropout): Dropout(p=0.1, inplace=False)
        (layers): ModuleList(
          (0-11): 12 x SpeechT5EncoderLayer(
            (attention): SpeechT5Attention(
              (k_proj): Linear(in_features=768, out_features=768, bias=True)
              (v_proj): Linear(in_features=768, out_features=768, bias=True)
              (q_proj): Linear(in_features=768, out_features=768, bias=True)
              (out_proj): Linear(in_features=768, out_features=768, bias=True)
            )
            (dropout): Dropout(p=0.1, inplace=False)
            (layer_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
            (feed_forward): SpeechT5FeedForward(
              (intermediate_dropout): Dropout(p=0.1, inplace=False)
              (intermediate_dense): Linear(in_features=768, out_features=3072, bias=True)
              (intermediate_act_fn): GELUActivation()
              (output_dense): Linear(in_features=3072, out_features=768, bias=True)
              (output_dropout): Dropout(p=0.1, inplace=False)
            )
            (final_layer_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          )
        )
        (embed_positions): SpeechT5RelativePositionalEncoding(
          (pe_k): Embedding(320, 64)
        )
      )
    )
    (decoder): SpeechT5DecoderWithTextPrenet(
      (prenet): SpeechT5TextDecoderPrenet(
        (dropout): Dropout(p=0.1, inplace=False)
        (embed_tokens): Embedding(81, 768, padding_idx=1)
        (embed_positions): SpeechT5SinusoidalPositionalEmbedding()
      )
      (wrapped_decoder): SpeechT5Decoder(
        (layers): ModuleList(
          (0-5): 6 x SpeechT5DecoderLayer(
            (self_attn): SpeechT5Attention(
              (k_proj): Linear(in_features=768, out_features=768, bias=True)
              (v_proj): Linear(in_features=768, out_features=768, bias=True)
              (q_proj): Linear(in_features=768, out_features=768, bias=True)
              (out_proj): Linear(in_features=768, out_features=768, bias=True)
            )
            (dropout): Dropout(p=0.1, inplace=False)
            (self_attn_layer_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
            (encoder_attn): SpeechT5Attention(
              (k_proj): Linear(in_features=768, out_features=768, bias=True)
              (v_proj): Linear(in_features=768, out_features=768, bias=True)
              (q_proj): Linear(in_features=768, out_features=768, bias=True)
              (out_proj): Linear(in_features=768, out_features=768, bias=True)
            )
            (encoder_attn_layer_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
            (feed_forward): SpeechT5FeedForward(
              (intermediate_dropout): Dropout(p=0.1, inplace=False)
              (intermediate_dense): Linear(in_features=768, out_features=3072, bias=True)
              (intermediate_act_fn): GELUActivation()
              (output_dense): Linear(in_features=3072, out_features=768, bias=True)
              (output_dropout): Dropout(p=0.1, inplace=False)
            )
            (final_layer_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          )
        )
      )
    )
  )
  (text_decoder_postnet): SpeechT5TextDecoderPostnet(
    (lm_head): Linear(in_features=768, out_features=81, bias=False)
  )
)
Downloads last month
30
Inference Examples
Text-to-Speech
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train Omarrran/quantized_english_speecht5_finetune-tts