Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
:computer: :robot: A summary on our attempts at using Deep Learning approaches for Emotional Text to Speech :speaker:
| Date | Stars |
|---|---|
| 2026-07-24 | 458 |
| 2026-07-25 | 458 |
| 2026-07-28 | 458 |
| 2026-07-30 | 458 |
| 2026-07-31 | 458 |
| 2026-08-06 | 458 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Deep Learning for Emotional Text-to-speech A summary on our attempts at using Deep Learning approaches for Emotional Text to Speech [](https://colab.research.google.com/github/Emotional-Text-to-Speech/dl-for-emo-tts/blob/master/Demo_DL_Based_Emotional_TTS.ipynb) [](https://zenodo.org/badge/latestdoi/260492782)  --- # Contents - [Datasets](#datasets) - [Relevant literature](#relevant-literature) - [Approach: Tacotron Models](#approach-tacotron-models) - [Approach 1: Fine-tuning a Vanilla Tacotron model on RAVDESS pre-trained on LJ Speech](#x-approach-1-fine-tuning-a-vanilla-tacotron-model-on-ravdess-pre-trained-on-lj-speech) - [Approach 2: Using a smaller learning rate for fine-tuning](#x-approach-2-using-a-smaller-learning-rate-for-fine-tuning) - [Approach 3: Using a smaller learning rate and SGD for fine-tuning](#x-approach-3-using-a-smaller-learning-rate-and-sgd-for-fine-tuning) - [Approach 4: Freezing the encoder and postnet](#x-approach-4-freezing-the-encoder-and-postnet) - [Approach 5: Freezing the encoder and postnet, and switching back to Adam](#x-approach-5-freezing-the-encoder-and-postnet-and-switching-back-to-adam) - [Approach 6: Freezing just the post-net, using Adam with low initial learning rate, training on EMOV-DB](#white_check_mark-approach-6-freezing-just-the-post-net-using-adam-with-low-initial-learning-rate-training-on-emov-db) - [Approach: DCTTS Models](#approach-dctts-models) - [Approach 7: Fine-tuning the Text2Mel module of the DC-TTS model on EMOV-DB pre-trained on LJ Speech](#x-approach-7-fine-tuning-the-text2mel-module-of-the-dc-tts-model-on-emov-db-pre-trained-on-lj-speech) - [Approach 8: Fine-tuning only on one speaker with reduced `top_db` and monotonic attention](#white_check_mark-approach-8-fine-tuning-only-on-one-speaker-with-reduced-top_db-and-monotonic-attention) - [Reproducibility and Code](#reproducibility-and-code) - [Demonstration](#demonstration) - [Cite](#cite) - [Contact](#contact) --- # Datasets | Dataset | No. of Speakers | Emotions | No. of utterances | No. of unique prompts | Duration | Language | Comments | Pros | Cons| | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | | **[RAVDESS](https://zenodo.org/record/1188976#.Xqw8ntMvPBI)** | 24 (12 female, 12 male) | 8 (calm, neutral, happy, sad, angry, fearful, surprise, and disgust) | 1440 | 2 | ~1 hour | English | <ul><li> Each speaker has 4 utterances for neutral emotion and 8 utterances for all other emotions, leading to 60 utterances per speaker</li></ul> | <ul><li>Easily available</li><li>Emotions contained are very easy to interpret</li></ul> | <ul><li>Very limited utterances</li><li>Poor vocabulary</li><li> Same utterance in different voices</li></ul> | | **[EMOV-DB](https://github.com/numediart/EmoV-DB)** | 5 (3 male, 2 female) | 5 (neutral, amused, angry sleepy, disgust) | 6914 (1568, 1315, 1293, 1720, 1018) | 1150 | ~7 hours | English, French (1 male speaker) |<ul><li>An attempt at a large scale corpus for Emotional speech</li><li> The *Amused* emotion contains non-verbal cues like chuckling, etc. which do not show up in the transcript</li><li> Similarly, *Sleepiness* has yawning sounds.</li></ul> | <ul><li>Only large scale emotional corpus that we found freely available</li></ul> | <ul><li>Emotions covered are not very easy to interpret</li><li>The non-verbal cues make synthsis difficult</li><li> Also, not all emotions are available for all speakers</li></ul> | | **[LJ Speech](https://keithito.com/LJ-Speech-Dataset/)** | 1 (1 female) | NA (can be considered neutral) | 13100 | 13100 | 23 hours 55 minutes 17 seconds | English | <ul><li>This is one of the largest corpuses for speech generation, with a rich vocabulary of over ~14k unique words</li><li>The sentences are taken from 7 non-fiction books</li></ul> | <ul><li>Large scale corpus</li><li> Rich vocab
Excerpt of 30,435 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:10ebe1ebf16cf1ad, topic:tts, topic:speech-synthesis, desc:text-to-speech