• Model the acoustic parameters of speech
• Synthesis audio from parameters
• Hidden Markov Model - precursor to tacotron

HMM based TTS pipeline

• Text
• Phonemes
• Duration Model (HMM)
• Acoustic Model (HMM)
• Vocoder - Parametric
• Waveform

graph TD

A[Text] -->|G2P| B[Phonemes]

B -->|State durations| C[Duration Model - HMM]

C -->|Timed states| D[Acoustic Model - HMM]

D -->|Spectral features| E[Vocoder]

E -->|Audio signal| F[Waveform]

Example

• HTS