forked from extern/Thorsten-Voice
Update for dataset version 2
This commit is contained in:
Thorsten Mueller
parent
b88f8a6260
commit
9bb90df9f7
23
README.md
23
README.md
@ -70,23 +70,22 @@ To get an impression what my voice sounds to decide if it fits to your project i
|
||||
> Interested in evolution of this dataset? See following pdf document ([evolution of thorsten dataset](./EvolutionOfThorstenDataset.pdf) )
|
||||
|
||||
## Download information
|
||||
> * https://drive.google.com/file/d/1yKJM1LAOQpRVojKunD9r8WN_p5KzBxjc/view?usp=sharing
|
||||
> * Download size: 2,7GB
|
||||
> Download size: 2,7GB
|
||||
|
||||
Version | Description | Link
|
||||
------------ | ------------- | ------------- | -------------
|
||||
thorsten-de-v01 | Initial version | [Google Drive Download v01](https://drive.google.com/file/d/1yKJM1LAOQpRVojKunD9r8WN_p5KzBxjc/view?usp=sharing)
|
||||
thorsten-de-v02 | normalized to -24dB and split metadata.csv into shuffeled metadata_train.csv and metadata_val.csv | [Google Drive Download v02](https://drive.google.com/file/d/1mGWfG0s2V2TEg-AI2m85tze1m4pyeM7b/view?usp=sharing)
|
||||
|
||||
|
||||
# Trained tacotron2 model "thorsten"
|
||||
> Training is currently in progress.
|
||||
If you trained a model on "thorsten" dataset please file an issue with some information on it. Sharing a trained model is highly appreciated.
|
||||
|
||||
> If you trained a model on "thorsten" dataset please file an issue with some information on it. Sharing a trained model is highly appreciated.
|
||||
## Trained models (TODO)
|
||||
|
||||
## Trained models (with at least acceptable) quality
|
||||
Inside the "models" (sub)folders are configs and Dockerfiles for a specific training from scratch.
|
||||
> Thanks to @erogol and @repodiac for brining in idea/code for script/container files.
|
||||
|
||||
|
||||
Folder | Date | Branch (Mozilla TTS repo) | Description
|
||||
Folder | Date | Link | Description
|
||||
------------ | ------------- | ------------- | -------------
|
||||
thorsten-taco2-v0.0.1 | august 2020 | dev | pure taco2 training without vocoder
|
||||
thorsten-taco2-v0.0.2 | to do | to do | to do
|
||||
thorsten-taco2-ddc-v0.1 | to do | to do | to do
|
||||
|
||||
# Feel free to file an issue if you ...
|
||||
* have improvements on dataset
|
||||
|
||||
@ -1,68 +0,0 @@
|
||||
# Adapted from @thorstenMueller's training script (https://github.com/thorstenMueller/TTS_recipes)
|
||||
# Docker file by @repodiac (https://github.com/repodiac/tit-for-tat/thorsten-tts)
|
||||
# *** Use without warranty and at your own risk! ***
|
||||
|
||||
# Installation folder **inside** Docker container
|
||||
# (NOTE: if it is changed to another folder, you have to manually change it to the same folder in the last line, ENTRYPOINT ...)
|
||||
ARG BASEDIR=/tmp/tts
|
||||
|
||||
FROM pytorch/pytorch:1.6.0-cuda10.1-cudnn7-devel as ttts-base
|
||||
ARG BASEDIR
|
||||
WORKDIR $BASEDIR
|
||||
|
||||
# Install system libraries etc.
|
||||
FROM ttts-base as ttts1
|
||||
ARG BASEDIR
|
||||
WORKDIR $BASEDIR
|
||||
|
||||
RUN apt-get update
|
||||
RUN apt-get install -y --no-install-recommends build-essential gcc espeak-ng espeak-ng-data git git-extras
|
||||
RUN pip install pip --upgrade
|
||||
RUN pip install gdown
|
||||
|
||||
# Clone deep-learning-german-tts repo and copy config and test sentences
|
||||
RUN git clone --single-branch --branch dev https://github.com/thorstenMueller/deep-learning-german-tts.git
|
||||
RUN cp $BASEDIR/deep-learning-german-tts/models/thorsten-taco2-v0.0.1/de-test-sentences.txt $BASEDIR
|
||||
RUN cp $BASEDIR/deep-learning-german-tts/models/thorsten-taco2-v0.0.1/config.json $BASEDIR
|
||||
|
||||
# Download and extract "thorsten-TTS" dataset
|
||||
FROM ttts1 as ttts2
|
||||
ARG BASEDIR
|
||||
WORKDIR $BASEDIR
|
||||
|
||||
RUN cd $BASEDIR
|
||||
RUN gdown https://drive.google.com/uc?id=1yKJM1LAOQpRVojKunD9r8WN_p5KzBxjc -O thorsten-dataset.tgz
|
||||
RUN tar -xvzf thorsten-dataset.tgz
|
||||
|
||||
# Prepare shuffled training and validate data (90% train, 10% val)
|
||||
RUN shuf LJSpeech-1.1/metadata.csv > LJSpeech-1.1/metadata_shuf.csv
|
||||
RUN head -n 20400 LJSpeech-1.1/metadata_shuf.csv > LJSpeech-1.1/metadata_train.csv
|
||||
RUN tail -n 2268 LJSpeech-1.1/metadata_shuf.csv > LJSpeech-1.1/metadata_val.csv
|
||||
|
||||
# Install Mozilla TTS repo and dependencies
|
||||
FROM ttts2 as ttts3
|
||||
ARG BASEDIR
|
||||
WORKDIR $BASEDIR
|
||||
|
||||
RUN git clone --single-branch --branch dev https://github.com/mozilla/TTS $BASEDIR/TTS
|
||||
|
||||
WORKDIR $BASEDIR/TTS
|
||||
RUN python setup.py develop
|
||||
|
||||
# Add german phoneme cleaner library by @repodiac
|
||||
FROM ttts3 as ttts4
|
||||
ARG BASEDIR
|
||||
|
||||
RUN git clone https://github.com/repodiac/german_transliterate $BASEDIR/german_transliterate
|
||||
WORKDIR $BASEDIR/german_transliterate
|
||||
RUN pip install -e .
|
||||
|
||||
WORKDIR $BASEDIR/TTS/mozilla_voice_tts/tts/utils/text
|
||||
RUN sed '/import re/a from german_transliterate.core import GermanTransliterate' cleaners.py >> cleaners-new.py
|
||||
RUN mv cleaners-new.py cleaners.py
|
||||
RUN echo "\ndef german_phoneme_cleaners(text):" >> cleaners.py
|
||||
RUN echo "\treturn GermanTransliterate(replace={';': ',', ':': ' '}, sep_abbreviation=' -- ').transliterate(text)" >> cleaners.py
|
||||
|
||||
# Run training
|
||||
WORKDIR $BASEDIR/TTS/mozilla_voice_tts/bin/
|
||||
ENTRYPOINT CUDA_VISIBLE_DEVICES="0" python train_tts.py --config_path $BASEDIR/config.json
|
||||
@ -1,3 +0,0 @@
|
||||
NOCH MIT REPODIAC KLÄREN WAS IN MEIN UND IN SEIN REPO KOMMT ODER WAS EINFACH VERLINKT WIRD
|
||||
|
||||
> https://github.com/repodiac/tit-for-tat/tree/master/thorsten-TTS
|
||||
@ -1,148 +0,0 @@
|
||||
{
|
||||
"github_branch":"* master",
|
||||
"restore_path":"",
|
||||
"model": "Tacotron2", // one of the model in models/
|
||||
"run_name": "thorsten-v1.0.0",
|
||||
"run_description": "thorsten-de v0.0.1",
|
||||
|
||||
// AUDIO PARAMETERS
|
||||
"audio":{
|
||||
// New "dev" branch params
|
||||
"fft_size": 1024,
|
||||
"spec_gain": 1.0,
|
||||
|
||||
// Audio processing parameters
|
||||
"num_mels": 80, // size of the mel spec frame.
|
||||
"num_freq": 1025, // number of stft frequency levels. Size of the linear spectogram frame.
|
||||
"sample_rate": 22050, // DATASET-RELATED: wav sample-rate. If different than the original data, it is resampled.
|
||||
"win_length": 1024, // stft window length in ms.
|
||||
"hop_length": 256, // stft window hop-lengh in ms.
|
||||
"frame_length_ms": null, // stft window length in ms.If null, 'win_length' is used.
|
||||
"frame_shift_ms": null, // stft window hop-lengh in ms. If null, 'hop_length' is used.
|
||||
"preemphasis": 0.0, //0.98, // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis.
|
||||
"min_level_db": -100, // normalization range
|
||||
"ref_level_db": 20, // reference level db, theoretically 20db is the sound of air.
|
||||
"power": 1.5, // value to sharpen wav signals after GL algorithm.
|
||||
"griffin_lim_iters": 60,// #griffin-lim iterations. 30-60 is a good range. Larger the value, slower the generation.
|
||||
// Normalization parameters
|
||||
"signal_norm": true, // normalize the spec values in range [0, 1]
|
||||
"symmetric_norm": true, // move normalization to range [-1, 1]
|
||||
"max_norm": 4.0, // scale normalization to range [-max_norm, max_norm] or [0, max_norm]
|
||||
"clip_norm": true, // clip normalized values into the range.
|
||||
"mel_fmin": 75.0, // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!!
|
||||
"mel_fmax": 7750.0, // maximum freq level for mel-spec. Tune for dataset!!
|
||||
"do_trim_silence": true, // enable trimming of slience of audio as you load it. LJspeech (false), TWEB (false), Nancy (true)
|
||||
"trim_db": 60 // threshold for timming silence. Set this according to your dataset.
|
||||
},
|
||||
|
||||
// VOCABULARY PARAMETERS
|
||||
// if custom character set is not defined,
|
||||
// default set in symbols.py is used
|
||||
"characters":{
|
||||
"pad": "_",
|
||||
"eos": "~",
|
||||
"bos": "^",
|
||||
"characters": "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz!'(),-.:;? äöüÄÖÜß",
|
||||
"punctuations":"!'(),-.:;? ",
|
||||
"phonemes":"iyɨʉɯuɪʏʊeøɘəɵɤoɛœɜɞʌɔæɐaɶɑɒᵻʘɓǀɗǃʄǂɠǁʛpbtdʈɖcɟkɡqɢʔɴŋɲɳnɱmʙrʀⱱɾɽɸβfvθðszʃʒʂʐçʝxɣχʁħʕhɦɬɮʋɹɻjɰlɭʎʟˈˌːˑʍwɥʜʢʡɕʑɺɧɚ˞ɫ"
|
||||
},
|
||||
|
||||
// DISTRIBUTED TRAINING
|
||||
"distributed":{
|
||||
"backend": "nccl",
|
||||
"url": "tcp:\/\/localhost:54321"
|
||||
},
|
||||
|
||||
"reinit_layers": [], // give a list of layer names to restore from the given checkpoint. If not defined, it reloads all heuristically matching layers.
|
||||
|
||||
// TRAINING
|
||||
"batch_size": 32, // Batch size for training. Lower values than 32 might cause hard to learn attention. It is overwritten by 'gradual_training'.
|
||||
"eval_batch_size":16,
|
||||
"r": 7, // Number of decoder frames to predict per iteration. Set the initial values if gradual training is enabled.
|
||||
"gradual_training": [[0, 7, 64], [1, 5, 64], [50000, 3, 32], [100000, 2, 32], [200000, 1, 32]], //set gradual training steps [first_step, r, batch_size]. If it is null, gradual training is disabled. For Tacotron, you might need to reduce the 'batch_size' as you proceeed.
|
||||
"loss_masking": true, // enable / disable loss masking against the sequence padding.
|
||||
"ga_alpha": 10.0, // weight for guided attention loss. If > 0, guided attention is enabled.
|
||||
"apex_amp_level": null,
|
||||
|
||||
// VALIDATION
|
||||
"run_eval": true,
|
||||
"test_delay_epochs": 10, //Until attention is aligned, testing only wastes computation time.
|
||||
"test_sentences_file": "/tmp/tts/de-test-sentences.txt", // set a file to load sentences to be used for testing. If it is null then we use default english sentences.
|
||||
|
||||
// OPTIMIZER
|
||||
"noam_schedule": false, // use noam warmup and lr schedule.
|
||||
"grad_clip": 1.0, // upper limit for gradients for clipping.
|
||||
"epochs": 1000, // total number of epochs to train.
|
||||
"lr": 0.0005, // Initial learning rate. If Noam decay is active, maximum learning rate.
|
||||
"wd": 0.000001, // Weight decay weight.
|
||||
"warmup_steps": 4000, // Noam decay steps to increase the learning rate from 0 to "lr"
|
||||
"seq_len_norm": false, // Normalize eash sample loss with its length to alleviate imbalanced datasets. Use it if your dataset is small or has skewed distribution of sequence lengths.
|
||||
|
||||
// TACOTRON PRENET
|
||||
"memory_size": -1, // ONLY TACOTRON - size of the memory queue used fro storing last decoder predictions for auto-regression. If < 0, memory queue is disabled and decoder only uses the last prediction frame.
|
||||
"prenet_type": "original", // "original" or "bn".
|
||||
"prenet_dropout": true, // enable/disable dropout at prenet.
|
||||
|
||||
// ATTENTION
|
||||
"attention_type": "original", // 'original' or 'graves'
|
||||
"attention_heads": 4, // number of attention heads (only for 'graves')
|
||||
"attention_norm": "softmax", // softmax or sigmoid. Suggested to use softmax for Tacotron2 and sigmoid for Tacotron.
|
||||
"windowing": false, // Enables attention windowing. Used only in eval mode.
|
||||
"use_forward_attn": true, // if it uses forward attention. In general, it aligns faster.
|
||||
"forward_attn_mask": false, // Additional masking forcing monotonicity only in eval mode.
|
||||
"transition_agent": false, // enable/disable transition agent of forward attention.
|
||||
"location_attn": true, // enable_disable location sensitive attention. It is enabled for TACOTRON by default.
|
||||
"bidirectional_decoder": false, // use https://arxiv.org/abs/1907.09006. Use it, if attention does not work well with your dataset.
|
||||
|
||||
"double_decoder_consistency": true,
|
||||
"ddc_r": 7,
|
||||
|
||||
|
||||
|
||||
// STOPNET
|
||||
"stopnet": true, // Train stopnet predicting the end of synthesis.
|
||||
"separate_stopnet": true, // Train stopnet seperately if 'stopnet==true'. It prevents stopnet loss to influence the rest of the model. It causes a better model, but it trains SLOWER.
|
||||
|
||||
// TENSORBOARD and LOGGING
|
||||
"print_step": 25, // Number of steps to log traning on console.
|
||||
"tb_plot_step": 100,
|
||||
"print_eval": false, // If True, it prints intermediate loss values in evalulation.
|
||||
"save_step": 5000, // Number of training steps expected to save traninpg stats and checkpoints.
|
||||
"checkpoint": true, // If true, it saves checkpoints per "save_step"
|
||||
"tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging.
|
||||
|
||||
// DATA LOADING
|
||||
//"text_cleaner": "phoneme_cleaners",
|
||||
"text_cleaner": "german_phoneme_cleaners",
|
||||
"enable_eos_bos_chars": false, // enable/disable beginning of sentence and end of sentence chars.
|
||||
"num_loader_workers": 8, // number of training data loader processes. Don't set it too big. 4-8 are good values.
|
||||
"num_val_loader_workers": 4, // number of evaluation data loader processes.
|
||||
"batch_group_size": 0, //Number of batches to shuffle after bucketing.
|
||||
"min_seq_len": 3, // DATASET-RELATED: minimum text length to use in training
|
||||
"max_seq_len": 180, // DATASET-RELATED: maximum text length
|
||||
|
||||
// PATHS
|
||||
"output_path": "/tmp/tts/models/thorsten/",
|
||||
|
||||
// PHONEMES
|
||||
"phoneme_cache_path": "mozilla_de_phonemes", // phoneme computation is slow, therefore, it caches results in the given folder.
|
||||
"use_phonemes": true, // use phonemes instead of raw characters. It is suggested for better pronounciation.
|
||||
"phoneme_language": "de", // depending on your target language, pick one from https://github.com/bootphon/phonemizer#languages
|
||||
|
||||
// MULTI-SPEAKER and GST
|
||||
"use_speaker_embedding": false, // use speaker embedding to enable multi-speaker learning.
|
||||
"style_wav_for_test": null, // path to style wav file to be used in TacotronGST inference.
|
||||
"use_gst": false, // TACOTRON ONLY: use global style tokens
|
||||
|
||||
// DATASETS
|
||||
"datasets": // List of datasets. They all merged and they get different speaker_ids.
|
||||
[
|
||||
{
|
||||
"name": "ljspeech",
|
||||
"path": "/tmp/tts/LJSpeech-1.1",
|
||||
"meta_file_train": "metadata_train.csv",
|
||||
"meta_file_val": "metadata_val.csv"
|
||||
}
|
||||
]
|
||||
|
||||
}
|
||||
@ -1,7 +0,0 @@
|
||||
Die aktuelle Außentemperatur beträgt sieben Grad Celsius und die Regenwahrscheinlichkeit liegt bei zwölf Prozent.
|
||||
Die aktuelle Außentemperatur beträgt 7°C und die Regenwahrscheinlichkeit liegt bei 12%.
|
||||
Frankfurt am Main wird auch Mainhattan genannt.
|
||||
Ich bedanke mich bei euch für euren Support und eure Gedult bei der Erzeugung einer künstlichen Stimme.
|
||||
Hallo, wie geht es Dir?
|
||||
Was ist los!
|
||||
Die wachsende Furcht vor den Folgen des grassierenden Coronavirus für die Weltwirtschaft hat den Dax am Dienstag auf das tiefste Niveau seit Oktober vergangenen Jahres gedrückt.
|
||||
Loading…
Reference in New Issue
Block a user