Tortoise-TTS setup with Conda and GPU troubleshooting

This tutorial, uploaded March 16, 2023, explains a Tortoise-TTS installation workflow with Conda, then demonstrates speech generation on a Lambda Labs cloud GPU instance. The speaker could not complete the setup on his M1 MacBook: he reports missing ARM dependency wheels and a failure to load the model on the CPU. The successful demonstration therefore uses a remote server rather than his own computer.

The walkthrough covers Miniconda installation, cloning the repository, creating a Python 3.9 environment and installing PyTorch for the machine's configuration. The demonstrated environment reports Tortoise 2.4.2. On the cloud instance, the speaker installs NumPy 1.23 before running the Python setup file to resolve a dependency mismatch. He later reinstalls PyTorch with the Linux, Conda and CUDA 11.7 selection after a driver or package error. These are details of the 2023 setup, not confirmation of current installation requirements.

For speech generation, he creates a Python script, loads the included Tom voice and uses the fast preset to save a WAV file. He does not demonstrate cloning his own voice. The sample takes roughly a minute to a minute and ten seconds on his instance; his suggestion that a local setup could be faster depends on the GPU and is not a benchmark.