Copy a token from your Hugging Face\ntokens page and paste it below. Immediately click login after copying\nyour token or it might be stored in plain text in this notebook file.
"
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"metadata": {
"tags": null
},
"name": "stderr",
"output_type": "stream",
"text": [
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n"
]
},
{
"output_type": "display_data",
"data": {
"text/plain": [
""
],
"text/html": [
"\n",
"
"
]
},
"metadata": {}
},
{
"output_type": "stream",
"name": "stderr",
"text": [
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n",
"/usr/local/lib/python3.10/dist-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.\n",
" warnings.warn(\n",
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n"
]
},
{
"output_type": "execute_result",
"data": {
"text/plain": [
"TrainOutput(global_step=10500, training_loss=0.4873873142969041, metrics={'train_runtime': 11266.2505, 'train_samples_per_second': 29.824, 'train_steps_per_second': 0.932, 'total_flos': 2.251261023307164e+16, 'train_loss': 0.4873873142969041, 'epoch': 90.51724137931035})"
]
},
"metadata": {},
"execution_count": 42
}
]
},
{
"cell_type": "markdown",
"source": [
"If we do one more `push_to_hub()` after training we can get a nice model card built for us. We simply have to set the appropriate keyword arguments (kwargs). You can change these values to match your dataset, language and model name accordingly:"
],
"metadata": {
"id": "4z3CzVGSs000"
}
},
{
"cell_type": "code",
"source": [
"kwargs = {\n",
" \"dataset_tags\": \"m-aliabbas/common_voice_urdu1\",\n",
" \"dataset\": \"common_voice_urdu1\", # a 'pretty' name for the training dataset\n",
" \"dataset_args\": \"split: train\",\n",
" \"model_name\": \"SpeechT5 TTS urdu\", # a 'pretty' name for your model\n",
" \"finetuned_from\": \"microsoft/speecht5_tts\",\n",
" \"tasks\": \"text-to-speech\",\n",
"}\n"
],
"metadata": {
"id": "NpXOJAZIs2-n"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "markdown",
"source": [
"The training results can now be uploaded to the Hub. To do so, execute the `push_to_hub` command:"
],
"metadata": {
"id": "ooEluBcXs5hJ"
}
},
{
"cell_type": "code",
"source": [
"trainer.push_to_hub(**kwargs)"
],
"metadata": {
"id": "mpodV89js9KT",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 156,
"referenced_widgets": [
"456aa13fa38c4c01b9f43570e31556fe",
"ec7ac27d32c44b40ac8fd18a7700962d",
"fbeb5a0f2394434fb67644f452dcb6b2",
"f22ab8b84c9a406d9383a1253576cf62",
"3915aa47172546afb3f09ea1f7c5b947",
"308b5dd72e6f4ea39d9be4c2e4a02512",
"9aa2c30314774289a458d92cabf2e9e1",
"4d5bdff9982e40138888bc9c72f9233e",
"2c9002098b114841b511b97c5fa39f14",
"fe75e4e2a67749b49efe29817c9148db",
"1660e3387e094116b87f3a8f04782124"
]
},
"outputId": "ecfd3656-aacf-4f48-f2bc-718285ecdcc6"
},
"execution_count": null,
"outputs": [
{
"output_type": "stream",
"name": "stderr",
"text": [
"Some non-default generation parameters are set in the model config. These should go into a GenerationConfig file (https://huggingface.co/docs/transformers/generation_strategies#save-a-custom-decoding-strategy-with-your-model) instead. This warning will be raised to an exception in v4.41.\n",
"Non-default generation parameters: {'max_length': 1876}\n",
"Your generation config was originally created from the model config, but the model config has changed since then. Unless you pass the `generation_config` argument to this model's `generate` calls, they will revert to the legacy behavior where the base `generate` parameterization is loaded from the model config instead. To avoid this behavior and this warning, we recommend you to overwrite the generation config model attribute before calling the model's `save_pretrained`, preferably also removing any generation kwargs from the model config. This warning will be raised to an exception in v4.41.\n"
]
},
{
"output_type": "display_data",
"data": {
"text/plain": [
"events.out.tfevents.1721676197.d34fde1d2e44.268.2: 0%| | 0.00/101k [00:00\u001b[0m in \u001b[0;36m\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msave_pretrained\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m 2\u001b[0m \u001b[0mtokenizer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msave_pretrained\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"your_model_save_directory\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0;32m/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py\u001b[0m in \u001b[0;36msave_pretrained\u001b[0;34m(self, save_directory, is_main_process, state_dict, save_function, push_to_hub, max_shard_size, safe_serialization, variant, token, save_peft_format, **kwargs)\u001b[0m\n\u001b[1;32m 2476\u001b[0m \u001b[0;32mreturn\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m 2477\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 2478\u001b[0;31m \u001b[0mos\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmakedirs\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msave_directory\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mexist_ok\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m 2479\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m 2480\u001b[0m \u001b[0;32mif\u001b[0m \u001b[0mpush_to_hub\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0;32m/usr/lib/python3.10/os.py\u001b[0m in \u001b[0;36mmakedirs\u001b[0;34m(name, mode, exist_ok)\u001b[0m\n\u001b[1;32m 223\u001b[0m \u001b[0;32mreturn\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m 224\u001b[0m \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 225\u001b[0;31m \u001b[0mmkdir\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmode\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m 226\u001b[0m \u001b[0;32mexcept\u001b[0m \u001b[0mOSError\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m 227\u001b[0m \u001b[0;31m# Cannot rely on checking for EEXIST, since the operating system\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0;31mFileNotFoundError\u001b[0m: [Errno 2] No such file or directory: ''"
]
}
]
},
{
"cell_type": "markdown",
"source": [
"You can now share this model with anyone using the link on the Hub."
],
"metadata": {
"id": "ojrUgMeEs_LP"
}
},
{
"cell_type": "markdown",
"source": [
"## Evaluate"
],
"metadata": {
"id": "5K1AQd8B7b-p"
}
},
{
"cell_type": "markdown",
"source": [
"After training finishes, let's use the model to synthesize some speech!\n",
"\n",
"I'm loading the model from the Hugging Face Hub, as the Colab notebook was terminated before training finished (which is why it's a good idea to use `push_to_hub=True` when training)."
],
"metadata": {
"id": "TrqUYy3-Qr1w"
}
},
{
"cell_type": "code",
"source": [
"model= SpeechT5ForTextToSpeech.from_pretrained(\"pocketmonkey/speecht5_tts_urdu\")"
],
"metadata": {
"id": "bos-fW6CHijQ",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 113,
"referenced_widgets": [
"af584652d0304a899e66314d61726323",
"3b74fdd0043949548e50b1d87413505b",
"958d0637ccad4af49815a2d137fee333",
"e10a5c2d674c4948978e0bc3fc2e8b0d",
"68a15f79b7ab42eaa0b14086f2b54ed6",
"6916935a59f94d6e95449be55732c80c",
"cef8c2e1224b4f0eb21e4963d53431b5",
"7597054472514485afa120d6266c91a5",
"6035b946f770456fbfd123adbb5e3be9",
"64eb31cf8d59468ea56a8e58c71d8122",
"0900daeba5cf4cb29d00a9c5ed5ad330",
"054b003700514935960c7ee0895b8c8e",
"2dcbcbfcbb3c455a8beee0d7f4c12c05",
"be44a4398328449daebb6bd48e2519f7",
"bcd4e9d178c944f6973b2d62fa1e5538",
"d8419ef35db5403c862e8321572ad376",
"af01cdbb16dd490984a749d2da6eff34",
"54783e715ad0408eaa512f7d28a374b2",
"d0f6e0eefec74d9aa2a0a00868c77a55",
"bd858f248a124f14833b044f7b0b71e5",
"d432e91ec63b4655a50bc775ee74aab4",
"af1e15c4f52e4ea5886d1631de2fbe39",
"a617fca4345c4d21a6c44d5ef8246f91",
"3109d5c2ffe04d5ebf296cc26a334a38",
"4d97d63c12ab4c37b53207cf91ac9794",
"a3f2bf1d001c4f2d8571d2a4db8972d1",
"e27fcac6c1be4ecc89028aa8368e6e89",
"c1b4ea7cfc0d434dbb68027af48a5f07",
"5e1fd59b51d1459da614ab5d51116d59",
"ced524c3f07b40e792db0bf1e9c3fe92",
"b257195cf3314cb4bc3aeddc4da142b9",
"95446dc9a4cc46469d1efa27c7896bfc",
"3cd2ea08c82e49a08396e31dfead7f6f"
]
},
"outputId": "ad9f4982-f54c-4faa-c710-f3f7df98d480"
},
"execution_count": null,
"outputs": [
{
"output_type": "display_data",
"data": {
"text/plain": [
"config.json: 0%| | 0.00/2.11k [00:00"
],
"image/png": "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\n"
},
"metadata": {}
}
]
},
{
"cell_type": "markdown",
"source": [
"Finally, use the vocoder to turn the spectrogram into sound."
],
"metadata": {
"id": "McG3EnpYRJVj"
}
},
{
"cell_type": "code",
"source": [
"with torch.no_grad():\n",
" speech = vocoder(spectrogram)"
],
"metadata": {
"id": "MpogdOe_7uoe"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"from IPython.display import Audio\n",
"Audio(speech.numpy(), rate=16000)"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 75
},
"id": "5H_FaVuW7wXt",
"outputId": "ce585e70-1e90-4cf5-af17-9ac311dee7ee"
},
"execution_count": null,
"outputs": [
{
"output_type": "execute_result",
"data": {
"text/plain": [
""
],
"text/html": [
"\n",
" \n",
" "
]
},
"metadata": {},
"execution_count": 61
}
]
},
{
"cell_type": "code",
"source": [
"import soundfile as sf\n",
"sf.write(\"output.wav\", speech.numpy(), samplerate=16000)"
],
"metadata": {
"id": "PV0nUOmO7xvv",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 183
},
"outputId": "3e0c913d-86ac-41e6-dba1-144238fa147d"
},
"execution_count": null,
"outputs": [
{
"output_type": "error",
"ename": "TypeError",
"evalue": "can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mTypeError\u001b[0m Traceback (most recent call last)",
"\u001b[0;32m\u001b[0m in \u001b[0;36m\u001b[0;34m()\u001b[0m\n\u001b[1;32m 1\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0msoundfile\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0msf\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0msf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwrite\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"output.wav\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mspeech\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnumpy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msamplerate\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m16000\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
"\u001b[0;31mTypeError\u001b[0m: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first."
]
}
]
},
{
"cell_type": "markdown",
"source": [],
"metadata": {
"id": "fiir-kCPYa60"
}
},
{
"cell_type": "markdown",
"source": [
"## Building a demo\n",
"\n",
"To showcase your newly fine-tuned model, make a demo on Hugging Face Spaces! We've created a [template Gradio demo](https://huggingface.co/spaces/Matthijs/speecht5-tts-demo) that you can easily copy and make your own.\n",
"\n",
"Click the link to duplicate the template demo to your account: https://huggingface.co/spaces/Matthijs/speecht5-tts-demo?duplicate=true (or from the three-dot button at the top choose **Duplicate this Space**).\n",
"\n",
"We recommend giving your space a similar name to your fine-tuned model (e.g. `speecht5_tts_voxpopuli_nl`) and setting the visibility to \"Public\".\n",
"\n",
"Once you've duplicated the Space to your account, click **Files and versions > app.py > edit**. Change the model identifier to your fine-tuned model (line 9). Scroll to the bottom of the page and click **Commit changes to main**. The demo will reboot, this time using your fine-tuned model.\n",
"\n",
"You can share this demo with your friends and family so that they can use the model that you've trained!"
],
"metadata": {
"id": "wm7B3zxrumfF"
}
},
{
"cell_type": "markdown",
"source": [
"## Conclusion\n",
"\n",
"That's it, you've fine-tuned SpeechT5 for TTS on a custom dataset!\n",
"\n",
"In our experience, it can be difficult to get good results out of this model. The results can be rather noisy and sometimes what the model outputs doesn't even sound like speech at all. A lot of this appears to be related to the speaker embeddings. Since SpeechT5 was pre-trained with English x-vectors, it gives the best results using those English speaker embeddings. So if the generated speech sounds bad, try using a different speaker embedding and it might improve.\n",
"\n",
"Of course, the demo above was only trained for 3000 iterations. Training for longer should improve the results. Even so, the speech clearly is Dutch instead of English, and it does capture the voice characteristics of the speaker (compare to the original audio in the example).\n",
"\n",
"Another thing to experiment with is the model's configuration. For example, try using `config.reduction_factor = 1` to see if this improves the results."
],
"metadata": {
"id": "W5_p2oWERU3C"
}
},
{
"cell_type": "markdown",
"source": [
"A final note on ethical concerns: While TTS technology has many beneficial uses, it can also be used for nefarious purposes such as imitating someone's voice without their permission. Please use TTS wisely and responsibly!"
],
"metadata": {
"id": "58ISNvOwuGiY"
}
},
{
"cell_type": "code",
"source": [],
"metadata": {
"id": "L-1ph4naugf6"
},
"execution_count": null,
"outputs": []
}
]
}