git clone https://github.com/neonbjb/tortoise-tts.git && cd tortoise-tts

    install_requires=[
        'tqdm',
        'rotary_embedding_torch',
        'inflect',
        'progressbar',
        'einops',
        'unidecode',
        'scipy',
        'librosa',
        'transformers==4.31.0',
        'tokenizers==0.11.1',    # 0.14.0 → 0.11.1
        'scikit-learn==1.6.1',    # 追加。stable版の最新にした
        'scipy==1.13.1'
        # 'deepspeed==0.8.3',
    ],

docker build . -t tortoise-tts

docker run --rm --gpus all \
    -e TORTOISE_MODELS_DIR=/models \
    -v ./user_data/models:/models \
    -v ./user_data/results:/results \
    -v $HOME/.cache/huggingface:/root/.cache/huggingface \
    -v .:/work \
    -it tortoise-tts

conda env list

# conda environments:
#
base                 * /root/miniconda
tortoise               /root/miniconda/envs/tortoise

conda activate tortoise

cd /work

python tortoise/do_tts.py --help

usage: do_tts.py [-h] [--text TEXT] [--voice VOICE] [--preset PRESET]
                 [--use_deepspeed USE_DEEPSPEED] [--kv_cache KV_CACHE] [--half HALF]
                 [--output_path OUTPUT_PATH] [--model_dir MODEL_DIR] [--candidates CANDIDATES]
                 [--seed SEED] [--produce_debug_state PRODUCE_DEBUG_STATE]
                 [--cvvp_amount CVVP_AMOUNT]

optional arguments:
  -h, --help            show this help message and exit
  --text TEXT           Text to speak.
  --voice VOICE         Selects the voice to use for generation. See options in voices/
                        directory (and add your own!) Use the & character to join two voices
                        together. Use a comma to perform inference on multiple voices.
  --preset PRESET       Which voice preset to use.
  --use_deepspeed USE_DEEPSPEED
                        Use deepspeed for speed bump.
  --kv_cache KV_CACHE   If you disable this please wait for a long a time to get the output
  --half HALF           float16(half) precision inference if True it's faster and take less vram
                        and ram
  --output_path OUTPUT_PATH
                        Where to store outputs.
  --model_dir MODEL_DIR
                        Where to find pretrained model checkpoints. Tortoise automatically
                        downloads these to .models, so thisshould only be specified if you have
                        custom checkpoints.
  --candidates CANDIDATES
                        How many output candidates to produce per-voice.
  --seed SEED           Random seed which can be used to reproduce results.
  --produce_debug_state PRODUCE_DEBUG_STATE
                        Whether or not to produce debug_state.pth, which can aid in reproducing
                        problems. Defaults to true.
  --cvvp_amount CVVP_AMOUNT
                        How much the CVVP model should influence the output.Increasing this can
                        in some cases reduce the likelihood of multiple speakers. Defaults to 0
                        (disabled)

python tortoise/do_tts.py --text "I'm going to speak this" --voice random --preset fast

ls -lt results/

total 296
-rw-r--r-- 1 root root 109134 May 19 16:34 random_0_1.wav
-rw-r--r-- 1 root root  91214 May 19 16:34 random_0_2.wav
-rw-r--r-- 1 root root  95822 May 19 16:34 random_0_0.wav