Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix weird pulse at the end of the model #276

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix weird pulse at the end of the model #276

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
155 changes: 81 additions & 74 deletions Colab/StyleTTS2_Demo_LibriTTS.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,8 +3,8 @@
{
"cell_type": "markdown",
"metadata": {
"id": "view-in-github",
"colab_type": "text"
"colab_type": "text",
"id": "view-in-github"
},
"source": [
"<a href=\"https://colab.research.google.com/github/yl4579/StyleTTS2/blob/main/Colab/StyleTTS2_Demo_LibriTTS.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
Expand All @@ -31,11 +31,9 @@
},
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [

]
"output_type": "stream",
"text": []
}
],
"source": [
Expand Down Expand Up @@ -237,6 +235,9 @@
" duration = torch.sigmoid(duration).sum(axis=-1)\n",
" pred_dur = torch.round(duration.squeeze()).clamp(min=1)\n",
"\n",
" # Eliminate potential noise at the end of the audio during generation.\n",
" if not text[-1].isalnum():\n",
" pred_dur[-1] = 1\n",
"\n",
" pred_aln_trg = torch.zeros(input_lengths, int(pred_dur.sum().data))\n",
" c_frame = 0\n",
Expand Down Expand Up @@ -314,6 +315,9 @@
" duration = torch.sigmoid(duration).sum(axis=-1)\n",
" pred_dur = torch.round(duration.squeeze()).clamp(min=1)\n",
"\n",
" # Eliminate potential noise at the end of the audio during generation.\n",
" if not text[-1].isalnum():\n",
" pred_dur[-1] = 1\n",
"\n",
" pred_aln_trg = torch.zeros(input_lengths, int(pred_dur.sum().data))\n",
" c_frame = 0\n",
Expand Down Expand Up @@ -397,6 +401,9 @@
" duration = torch.sigmoid(duration).sum(axis=-1)\n",
" pred_dur = torch.round(duration.squeeze()).clamp(min=1)\n",
"\n",
" # Eliminate potential noise at the end of the audio during generation.\n",
" if not text[-1].isalnum():\n",
" pred_dur[-1] = 1\n",
"\n",
" pred_aln_trg = torch.zeros(input_lengths, int(pred_dur.sum().data))\n",
" c_frame = 0\n",
Expand Down Expand Up @@ -994,16 +1001,21 @@
},
{
"cell_type": "markdown",
"metadata": {
"id": "hPKg9eYpL00f"
},
"source": [
"#### Extreme setting (`alpha = 1, beta=1`)\n",
"This setting uses 0% of the reference timbre and prosody and use the diffusion model to sample the entire style. This makes the speaker very dissimilar to the reference speaker."
],
"metadata": {
"id": "hPKg9eYpL00f"
}
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "Ei-7JOccL0bF"
},
"outputs": [],
"source": [
"# unseen speaker\n",
"path = \"Demo/reference_audio/1221-135767-0014.wav\"\n",
Expand All @@ -1013,25 +1025,25 @@
"for _ in range(5):\n",
" wav = inference(text, ref_s, diffusion_steps=10, alpha=1, beta=1, embedding_scale=1)\n",
" display(ipd.Audio(wav, rate=24000, normalize=False))"
],
"metadata": {
"id": "Ei-7JOccL0bF"
},
"execution_count": null,
"outputs": []
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "FVMPc3bhL3eL"
},
"source": [
"#### No variation (`alpha = 0, beta=0`)\n",
"This setting uses 100% of the reference timbre and prosody and do not use the diffusion model at all. This makes the speaker very similar to the reference speaker, but there is no variation."
],
"metadata": {
"id": "FVMPc3bhL3eL"
}
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "yh1QZ7uhL4wM"
},
"outputs": [],
"source": [
"# unseen speaker\n",
"path = \"Demo/reference_audio/1221-135767-0014.wav\"\n",
Expand All @@ -1041,35 +1053,35 @@
"for _ in range(5):\n",
" wav = inference(text, ref_s, diffusion_steps=10, alpha=0, beta=0, embedding_scale=1)\n",
" display(ipd.Audio(wav, rate=24000, normalize=False))"
],
"metadata": {
"id": "yh1QZ7uhL4wM"
},
"execution_count": null,
"outputs": []
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "T0EvkWrAMBDB"
},
"source": [
"### Extra fun!\n",
"\n",
"You can record your own voice and clone it using pre-trained StyleTTS 2 model here."
],
"metadata": {
"id": "T0EvkWrAMBDB"
}
]
},
{
"cell_type": "markdown",
"source": [
"#### Run the following cell to record your voice for 5 seconds. Please keep speaking to have the best effect."
],
"metadata": {
"id": "R985j5QONY8I"
}
},
"source": [
"#### Run the following cell to record your voice for 5 seconds. Please keep speaking to have the best effect."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "MWrFs0KWMBpz"
},
"outputs": [],
"source": [
"# all imports\n",
"from IPython.display import Javascript\n",
Expand Down Expand Up @@ -1106,71 +1118,71 @@
" with open('audio.wav','wb') as f:\n",
" f.write(b)\n",
" return 'audio.wav' # or webm ?"
],
"metadata": {
"id": "MWrFs0KWMBpz"
},
"execution_count": null,
"outputs": []
]
},
{
"cell_type": "markdown",
"source": [
"#### Please run this cell and speak:"
],
"metadata": {
"id": "z35qXwM0Nhx1"
}
},
"source": [
"#### Please run this cell and speak:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "KUEoFyQBMR-8"
},
"outputs": [],
"source": [
"print('Speak now for 5 seconds.')\n",
"audio = record(sec=5)\n",
"import IPython.display as ipd\n",
"display(ipd.Audio(audio, rate=24000, normalize=False))"
],
"metadata": {
"id": "KUEoFyQBMR-8"
},
"execution_count": null,
"outputs": []
]
},
{
"cell_type": "markdown",
"source": [
"#### Synthesize in your own voice"
],
"metadata": {
"id": "OQS_7IBpNmM1"
}
},
"source": [
"#### Synthesize in your own voice"
]
},
{
"cell_type": "code",
"source": [
"text = ''' StyleTTS 2 is a text to speech model that leverages style diffusion and adversarial training with large speech language models to achieve human level text to speech synthesis. ''' # @param {type:\"string\"}\n"
],
"execution_count": null,
"metadata": {
"cellView": "form",
"id": "c0I3LY7vM8Ta"
},
"execution_count": null,
"outputs": []
"outputs": [],
"source": [
"text = ''' StyleTTS 2 is a text to speech model that leverages style diffusion and adversarial training with large speech language models to achieve human level text to speech synthesis. ''' # @param {type:\"string\"}\n"
]
},
{
"cell_type": "code",
"source": [
"reference_dicts = {}\n",
"reference_dicts['You'] = audio"
],
"execution_count": null,
"metadata": {
"id": "80eW-pwxNCxu"
},
"execution_count": null,
"outputs": []
"outputs": [],
"source": [
"reference_dicts = {}\n",
"reference_dicts['You'] = audio"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "yIga6MTuNJaN"
},
"outputs": [],
"source": [
"start = time.time()\n",
"noise = torch.randn(1,1,256).to(device)\n",
Expand All @@ -1185,25 +1197,20 @@
" display(ipd.Audio(wav, rate=24000, normalize=False))\n",
" print('Reference:')\n",
" display(ipd.Audio(path, rate=24000, normalize=False))"
],
"metadata": {
"id": "yIga6MTuNJaN"
},
"execution_count": null,
"outputs": []
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"provenance": [],
"authorship_tag": "ABX9TyPQdFTqqVEknEG/ma/HMfU+",
"collapsed_sections": [
"aAGQPfgYIR23",
"eJdB_nCOIVIN",
"R985j5QONY8I"
],
"authorship_tag": "ABX9TyPQdFTqqVEknEG/ma/HMfU+",
"include_colab_link": true
"include_colab_link": true,
"provenance": []
},
"kernelspec": {
"display_name": "Python 3",
Expand Down
Loading