First draft for script/Dockerimage

README.md

@@ -78,6 +78,14 @@ To get an impression what my voice sounds to decide if it fits to your project i
 
 > 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 (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 of training | branch | description |
+|--------|------------------|-------------|---|---|
+|thorsten-taco2-v0.0.1|august 2020| dev | pure taco2 training without vocoder|
+|thorsten-taco2-v0.0.1|to do| to do | to do |
 
 # Feel free to file an issue if you ...
 * have improvements on dataset
@@ -85,7 +93,6 @@ To get an impression what my voice sounds to decide if it fits to your project i
 * want to share your trained "thorsten" model
 * get to know about any abuse usage of my voice
 
-
 # Special thanks
 I want to thank all open source communities for providing great projects.
 
@@ -111,6 +118,7 @@ Thank you Dominik (@domcross / https://github.com/domcross/)
 * https://github.com/MycroftAI/mimic-recording-studio
 * https://voice.mozilla.org/
 * https://github.com/mozilla/TTS
+(https://github.com/repodiac/tit-for-tat/tree/master/thorsten-TTS)
 * https://raw.githubusercontent.com/mozilla/voice-web/master/server/data/de/sentence-collector.txt
 
 We'll hear us in future :-)

models/thorsten-taco2-v0.0.1/Dockerfile

@@ -0,0 +1,68 @@
+# 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

models/thorsten-taco2-v0.0.1/config.json

@@ -0,0 +1,148 @@
+{
+    "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"
+                }
+            ]
+    
+    }

models/thorsten-taco2-v0.0.1/de-test-sentences.txt

@@ -0,0 +1,7 @@
+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.