AdaSpeechOptions
Options & Configuration · AiDotNet.TextToSpeech.Classic
Options for AdaSpeech (adaptive TTS with acoustic condition modeling and conditional layer normalization).
Inheritance: ModelOptions → TtsModelOptions → AcousticModelOptions → FastSpeech2Options → AdaSpeechOptions
Known derived types: AdaSpeech2Options
For Beginners
These options configure the AdaSpeech model. Default values follow the original paper settings.
How It Works
AdaSpeech “follows the basic structure in FastSpeech 2” and “other model configurations follow Ren et al. (2021)
unless otherwise stated” (Chen et al. 2021, §3), so these options extend FastSpeech2Options: 4 FFT
blocks in the phoneme encoder and in the mel decoder, hidden 256 (phoneme embedding, speaker embedding, attention
and FFN input/output), 2 heads, FFN filter 1024 with kernel 9, an 80-bin mel output. Audio is 16 kHz with a 12.5 ms
hop and 50 ms window. The acoustic condition encoders use 256 filters; the phoneme-level vectors are 4-dimensional.
Constructors
| Constructor | Summary |
|---|---|
AdaSpeechOptions() |
Creates the paper’s LibriTTS configuration. |
AdaSpeechOptions(AdaSpeechOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
double AcousticConditionDropout { get; set; } |
Gets or sets the dropout in the acoustic condition encoders and predictor. |
int AcousticConditionFilterSize { get; set; } |
Gets or sets the filters of the acoustic condition encoders and predictor (256). |
int NumSpeakers { get; set; } |
Gets or sets the number of speakers in the speaker embedding table (LibriTTS: 2456). |
int PhonemeConditionDim { get; set; } |
Gets or sets the dimension of the phoneme-level acoustic vectors (4). |
int PhonemeEncoderKernelSize { get; set; } |
Gets or sets the phoneme-level encoder’s and predictor’s convolution kernel (3). |
int UtteranceEncoderKernelSize { get; set; } |
Gets or sets the utterance-level encoder’s convolution kernel (5). |
int UtteranceEncoderStride { get; set; } |
Gets or sets the utterance-level encoder’s convolution stride (3). |