IAcousticModel<T>

Interfaces · AiDotNet.TextToSpeech.Interfaces

Interface for acoustic models that generate mel-spectrograms from text.

Implements: ITtsModel<T>

Known implementations: AcousticModelBase<T>, AdaSpeech<T>, AlignTTS<T>, CoMoSpeech<T>, DeepVoice3<T>, E2TTS<T>, F5TTS<T>, FastSpeech2<T>, FastSpeech<T>, ForwardTacotron<T>, GlowTTS<T>, GradTTS<T>, MatchaTTS<T>, NonAttentiveTacotron<T>, PortaSpeech<T>, ProDiff<T>, PromptTTS<T>, SpeedySpeech<T>, StyleTTS<T>, Tacotron2<T>, Tacotron<T>, TransformerTTS<T>, VoiceFlow<T>

How It Works

Acoustic models form the first stage of a two-stage TTS pipeline: Text -> [Acoustic Model] -> Mel-Spectrogram -> [Vocoder] -> Waveform. Architectures include:

  • Autoregressive: Tacotron, Tacotron 2 (attention-based seq2seq)
  • Non-autoregressive: FastSpeech, FastSpeech 2 (parallel with duration predictor)
  • Flow-based: Glow-TTS (monotonic alignment search + flow)
  • Diffusion-based: Grad-TTS (score-based diffusion decoder)

Properties

Property Summary
int FftSize { get; } Gets the FFT window size used for spectrogram computation.
int HopSize { get; } Gets the mel-spectrogram hop size in audio samples.
int MelChannels { get; } Gets the number of mel-spectrogram frequency channels (typically 80).

Methods

Method Summary
Tensor<T> TextToMel(string text) Generates a mel-spectrogram from text input.