microsoft

/

speecht5_vc

like 81

Seems that microsoft/speecht5_vc does not initializes all weights of SpeechT5ForSpeechToSpeech model.

Some weights of the model checkpoint at microsoft/speecht5_vc were not used when initializing SpeechT5ForSpeechToSpeech: ['speecht5.encoder.prenet.pos_conv_embed.conv.weight_g', 'speecht5.encoder.prenet.pos_conv_embed.conv.weight_v']
- This IS expected if you are initializing SpeechT5ForSpeechToSpeech from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing SpeechT5ForSpeechToSpeech from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of SpeechT5ForSpeechToSpeech were not initialized from the model checkpoint at microsoft/speecht5_vc and are newly initialized: ['speecht5.encoder.prenet.pos_conv_embed.conv.parametrizations.weight.original0', 'speecht5.encoder.prenet.pos_conv_embed.conv.parametrizations.weight.original1', 'speecht5.encoder.prenet.pos_sinusoidal_embed.weights']