SparkTTS Voice cloning (Wav2vec)

[Blaizzy]

• Add Wav2vec model as STT (ASR and Audio classification)
• The torch mel seems to work better than current
• Fix model skipping the first few words