Catalan Text To Speech
Published:
This Repo Contains Implementation of Catalan Text To Speech With Some Samples
Based on Microsoft’s FastSpeech
Catalan Version of FastSpeech Repo
The model has following advantages:
- Robustness: No repeats and failed attention modes for challenging sentences.
- Speed: The generation of a mel spectogram takes about 0.04s on a GeForce RTX 2080.
- Controllability: It is possible to control the speed of the generated utterance.
- Efficiency: In contrast to FastSpeech and Tacotron, the model of ForwardTacotron does not use any attention. Hence, the required memory grows linearly with text size, which makes it possible to synthesize large articles at once.
Check out the latest audio samples (ForwardTacotron + WaveRNN)! The samples are generated with a model trained on 2 hoours of data from Catalan Common Voice and vocoded with WaveRNN, MelGAN, or HiFiGAN. You can try out the latest pretrained model with the following notebook:
Samples 🔉
Comparing Ground Truth voices with model generated voices.
| Text | Ground Truth | WaveRNN |
|---|---|---|
| És un edifici entre mitgeres fent xamfrà | ||
| Igualment, el propietari no va acceptar vendre | ||
| La resta de la vora és aproximadament circular | ||
| Només s’ha descrit a la seva localitat tipus |
More model generated voices with Catalan Text To Speech (Long and challenging sentences)
| Text | WaveRNN |
|---|---|
| L’àliga no caça mosques. | |
| Entrar per una orella i sortir per l’altre. | |
| La paraula bona molt val i poc costa. | |
| Quan el gat no hi es, les rates ballen. | |
| Val més mala avinença que bona sentencia. | |
| El president Trump es va reunir amb altres líders a la conferència del Grup dels 20. | |
| De mica en mica s’omple la pica i de gota en gota s’omple la bota. | |
| Científics del laboratori del CERN diuen que han descobert una nova partícula. | |
| La normativa del català és establerta, d’una manera general, per l’Institut d’Estudis Catalans, que pren com a base l’ortografia, la gramàtica i el diccionari elaborats per Pompeu Fabra, i per a les variants específiques del valencià, per l’Acadèmia Valenciana de la Llengua, que pren com a base les Normes de Castelló, és a dir, l’ortografia de Pompeu Fabra però més adaptada a la pronunciació del català occidental i als trets que caracteritzen les varietats valencianes. |
Speed controlibility of Catalan Text To Speech with different factors
| Text | Normal (Speed = 1) | Fast(Speed = 1.5) | Slow(Speed = .5) |
|---|---|---|---|
| Les persones es troben, les muntanyes no. | |||
| La paciència és la mare de la ciència | |||
| L’àliga no caça mosques |
References
- FastSpeech: Fast, Robust and Controllable Text to Speech
- FastPitch: Parallel Text-to-speech with Pitch Prediction
- HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis
- MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesis
Acknowlegements
- https://github.com/keithito/tacotron
- https://github.com/fatchord/WaveRNN
- https://github.com/seungwonpark/melgan
- https://github.com/jik876/hifi-gan
- https://github.com/xcmyz/LightSpeech
- https://github.com/resemble-ai/Resemblyzer
- https://github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/SpeechSynthesis/FastPitch
Maintainers
- Mehdi Hosseini Moghadam, github: mehdihosseinimoghadam
Copyright
See LICENSE for details.