diff --git a/README-CN.md b/README-CN.md index ea650cf..5ae338a 100644 --- a/README-CN.md +++ b/README-CN.md @@ -8,6 +8,11 @@ ## šŸ“£ ꛓꖰ +[2025-05-14]āš’ļø: ę”Æꌁ v1.5 ē‰ˆęœ¬. ęؔ型äø‹č½½å¹¶ę›“åę”¾åˆ° `ComfyUI\models\TTS\Index-TTS` č·Æ径äø‹: +- https://huggingface.co/IndexTeam/IndexTTS-1.5/blob/main/bigvgan_generator.pth ā†’ `bigvgan_generator_v1_5.pth` +- https://huggingface.co/IndexTeam/IndexTTS-1.5/blob/main/bpe.model ā†’ `bpe_v1_5.model` +- https://huggingface.co/IndexTeam/IndexTTS-1.5/blob/main/gpt.pth ā†’ `gpt_v1_5.pth` + [2025-05-02]āš’ļø: åÆē”Ø DeepSpeed 加速, éœ€č¦å®‰č£… DeepSpeed, Windows čÆ¦č§ [DeepSpeed å®‰č£…](https://github.com/deepspeedai/DeepSpeed/blob/master/blogs/windows/08-2024/chinese/README.md). 加速äøę˜Žę˜¾. [2025-04-30]āš’ļø: 发åøƒ v1.0.0. diff --git a/README.md b/README.md index 3876d10..8a9a44b 100644 --- a/README.md +++ b/README.md @@ -6,6 +6,11 @@ High-quality voice cloning, very fast, supports Chinese and English, and allows ## šŸ“£ Updates +[2025-05-14]āš’ļø: Supports v1.5. Download the models and rename they, placed in the `ComfyUI\models\TTS\Index-TTS` path. +- https://huggingface.co/IndexTeam/IndexTTS-1.5/blob/main/bigvgan_generator.pth ā†’ `bigvgan_generator_v1_5.pth` +- https://huggingface.co/IndexTeam/IndexTTS-1.5/blob/main/bpe.model ā†’ `bpe_v1_5.model` +- https://huggingface.co/IndexTeam/IndexTTS-1.5/blob/main/gpt.pth ā†’ `gpt_v1_5.pth` + [2025-05-02] āš’ļø: DeepSpeed acceleration is available, but DeepSpeed needs to be installed. For Windows, please refer to [DeepSpeed Installation](https://github.com/deepspeedai/DeepSpeed/blob/master/blogs/windows/08-2024/chinese/README.md). The acceleration is not obvious. [2025-04-30] āš’ļø: Released v1.0.0. diff --git a/checkpoints/config_v1_5.yaml b/checkpoints/config_v1_5.yaml new file mode 100644 index 0000000..ce63322 --- /dev/null +++ b/checkpoints/config_v1_5.yaml @@ -0,0 +1,112 @@ +dataset: + bpe_model: bpe_v1_5.model + sample_rate: 24000 + squeeze: false + mel: + sample_rate: 24000 + n_fft: 1024 + hop_length: 256 + win_length: 1024 + n_mels: 100 + mel_fmin: 0 + normalize: false + +gpt: + model_dim: 1280 + max_mel_tokens: 800 + max_text_tokens: 600 + heads: 20 + use_mel_codes_as_input: true + mel_length_compression: 1024 + layers: 24 + number_text_tokens: 12000 + number_mel_codes: 8194 + start_mel_token: 8192 + stop_mel_token: 8193 + start_text_token: 0 + stop_text_token: 1 + train_solo_embeddings: false + condition_type: "conformer_perceiver" + condition_module: + output_size: 512 + linear_units: 2048 + attention_heads: 8 + num_blocks: 6 + input_layer: "conv2d2" + perceiver_mult: 2 + +vqvae: + channels: 100 + num_tokens: 8192 + hidden_dim: 512 + num_resnet_blocks: 3 + codebook_dim: 512 + num_layers: 2 + positional_dims: 1 + kernel_size: 3 + smooth_l1_loss: true + use_transposed_convs: false + +bigvgan: + adam_b1: 0.8 + adam_b2: 0.99 + lr_decay: 0.999998 + seed: 1234 + + resblock: "1" + upsample_rates: [4,4,4,4,2,2] + upsample_kernel_sizes: [8,8,4,4,4,4] + upsample_initial_channel: 1536 + resblock_kernel_sizes: [3,7,11] + resblock_dilation_sizes: [[1,3,5], [1,3,5], [1,3,5]] + feat_upsample: false + speaker_embedding_dim: 512 + cond_d_vector_in_each_upsampling_layer: true + + gpt_dim: 1280 + + activation: "snakebeta" + snake_logscale: true + + use_cqtd_instead_of_mrd: true + cqtd_filters: 128 + cqtd_max_filters: 1024 + cqtd_filters_scale: 1 + cqtd_dilations: [1, 2, 4] + cqtd_hop_lengths: [512, 256, 256] + cqtd_n_octaves: [9, 9, 9] + cqtd_bins_per_octaves: [24, 36, 48] + + resolutions: [[1024, 120, 600], [2048, 240, 1200], [512, 50, 240]] + mpd_reshapes: [2, 3, 5, 7, 11] + use_spectral_norm: false + discriminator_channel_mult: 1 + + use_multiscale_melloss: true + lambda_melloss: 15 + + clip_grad_norm: 1000 + + segment_size: 16384 + num_mels: 100 + num_freq: 1025 + n_fft: 1024 + hop_size: 256 + win_size: 1024 + + sampling_rate: 24000 + + fmin: 0 + fmax: null + fmax_for_loss: null + mel_type: "pytorch" + + num_workers: 2 + dist_config: + dist_backend: "nccl" + dist_url: "tcp://localhost:54321" + world_size: 1 + +dvae_checkpoint: dvae.pth +gpt_checkpoint: gpt_v1_5.pth +bigvgan_checkpoint: bigvgan_generator_v1_5.pth diff --git a/indexttsnode.py b/indexttsnode.py index 1b0c587..f48783e 100644 --- a/indexttsnode.py +++ b/indexttsnode.py @@ -507,6 +507,7 @@ class IndexTTSRun: def INPUT_TYPES(s): return { "required": { + "version":(["v1.5", "V1.0"], {"default": "v1.5"}), "audio_prompt":("AUDIO",), "text": ("STRING", {"forceInput": True}), "text_language": (["zh", "en"], {"default": "zh"}), @@ -526,6 +527,7 @@ class IndexTTSRun: CATEGORY = "šŸŽ¤MW/MW-IndexTTS" def clone(self, + version, audio_prompt, text, text_language, @@ -537,8 +539,13 @@ class IndexTTSRun: fast_inference=True, unload_model=True ): + if version == "v1.5": + cfg_path=f"{current_dir}/checkpoints/config_v1_5.yaml" + else: + cfg_path=f"{current_dir}/checkpoints/config.yaml" + if self.index_tts is None: - self.index_tts = IndexTTS(text_language=text_language) + self.index_tts = IndexTTS(cfg_path=cfg_path, text_language=text_language) if fast_inference: res = self.index_tts.infer_fast( diff --git a/pyproject.toml b/pyproject.toml index 6e13bdf..5306337 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -1,7 +1,7 @@ [project] name = "indextts-mw" description = "IndexTTS Voice Cloning Nodes for ComfyUI. High-quality voice cloning, very fast, supports Chinese and English, and allows custom voice styles." -version = "1.0.2" +version = "1.1.0" license = {file = "LICENSE"} dependencies = ["# accelerate==0.25.0", "# transformers==4.36.2", "# tokenizers==0.15.0", "# cn2an==0.5.22", "# ffmpeg-python==0.2.0", "# Cython==3.0.7", "# g2p-en==2.1.0", "# jieba==0.42.1", "# keras==2.9.0", "# numba==0.58.1", "# numpy==1.26.2", "# pandas==2.1.3", "# matplotlib==3.8.2", "# opencv-python==4.9.0.80", "# vocos==0.1.0", "# accelerate==0.25.0", "# tensorboard==2.9.1", "omegaconf", "sentencepiece", "librosa", "tqdm", "# deepspeeds # Use it to accelerate model inference"]