sinapsis-speech 0.1.0__py3-none-any.whl → 0.2.0__py3-none-any.whl

{sinapsis_speech-0.1.0.dist-info → sinapsis_speech-0.2.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sinapsis-speech
-Version: 0.1.0
+Version: 0.2.0
 Summary: Generate speech using various libraries.
 Author-email: SinapsisAI <dev@sinapsis-ai.com>
 License:                     GNU AFFERO GENERAL PUBLIC LICENSE
@@ -666,25 +666,20 @@ License:                     GNU AFFERO GENERAL PUBLIC LICENSE
         <https://www.gnu.org/licenses/>.
         
 Project-URL: Homepage, https://sinapsis.tech
-Project-URL: Documentation, https://docs.sinapsis.tech/docs
+Project-URL: Documentation, https://docs.sinapsis.tech/docs/sinapsis-speech
 Project-URL: Tutorials, https://docs.sinapsis.tech/tutorials
 Project-URL: Repository, https://github.com/Sinapsis-AI/sinapsis-speech.git
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
-License-File: LICENSE
 Requires-Dist: pip>=24.3.1
-Requires-Dist: sinapsis>=0.1.1
-Provides-Extra: elevenlabs-app
-Requires-Dist: sinapsis-elevenlabs; extra == "elevenlabs-app"
-Requires-Dist: sinapsis-speech[gradio-app]; extra == "elevenlabs-app"
-Provides-Extra: gradio-app
-Requires-Dist: gradio>=5.14.0; extra == "gradio-app"
-Requires-Dist: sinapsis-data-readers>=0.1.0; extra == "gradio-app"
+Requires-Dist: sinapsis>=0.2.2
 Provides-Extra: all
-Requires-Dist: sinapsis-elevenlabs; extra == "all"
-Requires-Dist: sinapsis-speech[gradio-app]; extra == "all"
-Requires-Dist: sinapsis-speech[elevenlabs-app]; extra == "all"
-Dynamic: license-file
+Requires-Dist: sinapsis-elevenlabs[all]; extra == "all"
+Requires-Dist: sinapsis-f5-tts[all]; extra == "all"
+Requires-Dist: sinapsis-speech[webapp]; extra == "all"
+Requires-Dist: sinapsis-zonos[all]; extra == "all"
+Provides-Extra: gradio-app
+Requires-Dist: sinapsis[webapp]>=0.2.3; extra == "gradio-app"
 
 <h1 align="center">
 <br>
@@ -702,7 +697,7 @@ Sinapsis Speech
 <p align="center">
 <a href="#installation">🐍 Installation</a> •
 <a href="#packages">📦 Packages</a> •
-<a href="#webapp">🌐 Webapp</a> •
+<a href="#webapp">🌐 Webapps</a> •
 <a href="#documentation">📙 Documentation</a> •
 <a href="#packages">🔍 License</a>
 </p>
@@ -715,47 +710,93 @@ Sinapsis Speech
 > Sinapsis projects requires Python 3.10 or higher.
 >
 
-We strongly encourage the use of <code>uv</code>, although any other package manager should work too.
-If you need to install <code>uv</code> please see the [official documentation](https://docs.astral.sh/uv/getting-started/installation/#installation-methods).
+This repo includes packages for performing speech synthesis using different tools:
 
+* <code>sinapsis-elevenlabs</code>
+* <code>sinapsis-f5-tts</code>
+* <code>sinapsis-zonos</code>
 
-1. Install using your favourite package manager.
+Install using your preferred package manager. We strongly recommend using <code>uv</code>. To install <code>uv</code>, refer to the [official documentation](https://docs.astral.sh/uv/getting-started/installation/#installation-methods).
 
-Example with <code>uv</code>:
+
+Install with <code>uv</code>:
 ```bash
-  uv pip install sinapsis-elevenlabs --extra-index-url https://pypi.sinapsis.tech
+uv pip install sinapsis-elevenlabs --extra-index-url https://pypi.sinapsis.tech
 ```
- or with raw <code>pip</code>:
+Or with raw <code>pip</code>:
 ```bash
-  pip install sinapsis-elevenlabs --extra-index-url https://pypi.sinapsis.tech
+pip install sinapsis-elevenlabs --extra-index-url https://pypi.sinapsis.tech
 ```
-**Change the name of the package for the one you want to install**.
+
+**Replace `sinapsis-elevenlabs` with the name of the package you intend to install**.
+
+> [!IMPORTANT]
+> Templates in each package may require additional dependencies. For development, we recommend installing the package all optional dependencies:
+>
+With <code>uv</code>:
+
+```bash
+uv pip install sinapsis-elevenlabs[all] --extra-index-url https://pypi.sinapsis.tech
+```
+Or with raw <code>pip</code>:
+```bash
+pip install sinapsis-elevenlabs[all] --extra-index-url https://pypi.sinapsis.tech
+```
+
+**Be sure to substitute `sinapsis-elevenlabs` with the appropriate package name**.
+
+
 
 > [!TIP]
 > You can also install all the packages within this project:
 >
 ```bash
-  uv pip install sinapsis-speech[all] --extra-index-url https://pypi.sinapsis.tech
+uv pip install sinapsis-speech[all] --extra-index-url https://pypi.sinapsis.tech
 ```
 
 
 <h2 id="packages">📦 Packages</h2>
 
-Each package can be used independently or combined to create more complex workflows. Below is an overview of the available packages:
+This repository is organized into modular packages, each designed for integration with different text-to-speech tools. These packages provide ready-to-use templates for speech synthesis. Below is an overview of the available packages:
 
 <details>
-<summary id="elevenlabs"><strong><span style="font-size: 1.4em;"> Elevenlabs </span></strong></summary>
+<summary id="elevenlabs"><strong><span style="font-size: 1.4em;"> Sinapsis ElevenLabs </span></strong></summary>
+
+This package offers a suite of templates and utilities designed for effortless integrating, configuration, and execution of **text-to-speech (TTS)** and **voice generation** functionalities powered by [ElevenLabs](https://elevenlabs.io/).
 
-This package provides a suite of templates and utilities for seamlessly integrating, configuring, and running **text-to-speech (TTS)** and **voice generation** functionalities powered by [ElevenLabs](https://elevenlabs.io/):
+- **ElevenLabsTTS**: Template for converting text into speech using ElevenLabs' voice models.
 
-- **Text-to-speech**: Template for converting text into speech using ElevenLabs' voice models.
+- **ElevenLabsVoiceGeneration**: Template for generating custom synthetic voices based on user-provided descriptions.
 
-- **Voice generation**: Template for generating custom synthetic voices based on user-provided descriptions.
+For specific instructions and further details, see the [README.md](https://github.com/Sinapsis-AI/sinapsis-speech/blob/main/packages/sinapsis_elevenlabs/README.md).
 
 </details>
-<h2 id="webapps">🌐 Webapps</h2>
-The webapps included in this project showcase the modularity of the templates, in this case
-for speech generation tasks.
+
+
+<details>
+<summary id="f5tts"><strong><span style="font-size: 1.4em;"> Sinapsis F5-TTS</span></strong></summary>
+
+This package provides a template for seamlessly integrating, configuring, and running **text-to-speech (TTS)** functionalities powered by [F5TTS](https://github.com/SWivid/F5-TTS).
+
+- **F5TTSInference**: Converts text to speech using the F5TTS model with voice cloning capabilities.
+
+For specific instructions and further details, see the [README.md](https://github.com/Sinapsis-AI/sinapsis-speech/blob/main/packages/sinapsis_f5_tts/README.md).
+
+</details>
+
+<details>
+<summary id="zonos"><strong><span style="font-size: 1.4em;"> Sinapsis Zonos</span></strong></summary>
+
+This package provides a single template for integrating, configuring, and running **text-to-speech (TTS)** and **voice cloning** functionalities powered by [Zonos](https://github.com/Zyphra/Zonos/tree/main).
+
+- **ZonosTTS**: Template for converting text to speech or performing voice cloning based on the presence of an audio sample.​
+
+For specific instructions and further details, see the [README.md](https://github.com/Sinapsis-AI/sinapsis-speech/blob/main/packages/sinapsis_zonos/README.md).
+
+</details>
+
+<h2 id="webapp">🌐 Webapps</h2>
+The webapps included in this project showcase the modularity of the templates, in this case for speech generation tasks.
 
 > [!IMPORTANT]
 > To run the app you first need to clone this repository:
@@ -768,89 +809,102 @@ cd sinapsis-speech
 > [!NOTE]
 > If you'd like to enable external app sharing in Gradio, `export GRADIO_SHARE_APP=True`
 
-> [!IMPORTANT]
-> The CosyVoice model requires at least 4GB of ram to work.
 
 > [!IMPORTANT]
-> Elevenlabs requires an api key to run any inference. Please go to the [official website](https://elevenlabs.io), create an account.
-If you already have an account, go to the [token page](https://elevenlabs.io/app/settings/api-keys) and generate a token.
+> Elevenlabs requires an API key to run any inference. To get started, visit the [official website](https://elevenlabs.io) and create an account. If you already have an account, go to the [API keys page](https://elevenlabs.io/app/settings/api-keys) to generate a token.
 
 > [!IMPORTANT]
-> set your env var using <code> export ELEVENLABS_API_KEY='your-api-key'</code>
+> Set your env var using <code> export ELEVENLABS_API_KEY='your-api-key'</code>
 
+> [!IMPORTANT]
+> F5-TTS requires a reference audio file for voice cloning. Make sure you have a reference audio file in the artifacts directory.
 
-> [!TIP]
-> The agent configuration can be updated using the AGENT_CONFIG_PATH environment var.
+> [!NOTE]
+> Agent configuration can be changed through the `AGENT_CONFIG_PATH` env var. You can check the available configurations in each package configs folder.
 
 
 <details>
-<summary id="docker"><strong><span style="font-size: 1.4em;">🐳 Build with Docker</span></strong></summary>
+<summary id="docker"><strong><span style="font-size: 1.4em;">🐳 Docker</span></strong></summary>
 
-**IMPORTANT** This docker image depends on the sinapsis-nvidia:base image. Please refer to the official [sinapsis](https://github.com/Sinapsis-ai/sinapsis?tab=readme-ov-file#docker) instructions to Build with Docker.
+**IMPORTANT**: This Docker image depends on the `sinapsis-nvidia:base` image. For detailed instructions, please refer to the [Sinapsis README](https://github.com/Sinapsis-ai/sinapsis?tab=readme-ov-file#docker).
 
-1. **Build the Docker image**:
+1. **Build the sinapsis-speech image**:
 ```bash
 docker compose -f docker/compose.yaml build
 ```
 
-
-2. **Launch the service**:
+2. **Start the app container**:
+For ElevenLabs:
 ```bash
 docker compose -f docker/compose_apps.yaml up -d sinapsis-elevenlabs
 ```
+For F5-TTS:
+```bash
+docker compose -f docker/compose_apps.yaml up -d sinapsis-f5_tts
+```
+For Zonos:
+```bash
+docker compose -f docker/compose_apps.yaml up -d sinapsis-zonos
+```
 
-
-2. **Check the logs**
+3. **Check the logs**
+For ElevenLabs:
 ```bash
 docker logs -f sinapsis-elevenlabs
 ```
-3. **The logs will display the URL to access the webapp, e.g.,:**:
+For F5-TTS:
+```bash
+docker logs -f sinapsis-f5tts
+```
+For Zonos:
+```bash
+docker logs -f sinapsis-zonos
+```
+4. **The logs will display the URL to access the webapp, e.g.,:**:
 ```bash
 Running on local URL:  http://127.0.0.1:7860
 ```
-4. To stop the app:
+**NOTE**: The url may be different, check the output of logs.
+5. **To stop the app**:
 ```bash
-docker compose -f docker/compose_apps.yaml down sinapsis-elevenlabs
+docker compose -f docker/compose_apps.yaml down
 ```
 </details>
 
 <details>
 <summary id="virtual-environment"><strong><span style="font-size: 1.4em;">💻 UV</span></strong></summary>
 
+To run the webapp using the <code>uv</code> package manager, follow these steps:
 
 1. **Sync the virtual environment**:
 
 ```bash
 uv sync --frozen
 ```
-2. Install the wheel:
+2. **Install the wheel**:
 
 ```bash
 uv pip install sinapsis-speech[all] --extra-index-url https://pypi.sinapsis.tech
 ```
 
-
-3. **Activate the virtual environment**:
-
+3. **Run the webapp**:
+For ElevenLabs:
 ```bash
-source .venv/bin/activate
+uv run webapps/elevenlabs/elevenlabs_tts_app.py
 ```
-4. **Declare PYTHONPATH**
+For F5-TTS:
 ```bash
-export PYTHONPATH=$PWD/webapps 
+uv run webapps/f5-tts/f5_tts_app.py
 ```
-**NOTE** if not located in <code>sinapsis-speech</code> folder, change $PWD for the actual path to <code>sinapsis-speech</code>
-
-5. **Launch the demo**:
-
+For Zonos:
 ```bash
-python webapps/elevenlabs/elevenlabs_tts_app.py
+uv run webapps/zonos/zonos_tts_app.py
 ```
-6. Open the displayed URL, e.g.:
+4. **The terminal will display the URL to access the webapp (e.g.)**:
 ```bash
 Running on local URL:  http://127.0.0.1:7860
 ```
-**NOTE**: The URL can be different, please make sure you check the logs.
+**NOTE**: The URL may vary; check the terminal output for the correct address.
 
 </details>
 

sinapsis_speech-0.2.0.dist-info/RECORD

@@ -0,0 +1,21 @@
+sinapsis_elevenlabs/src/sinapsis_elevenlabs/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sinapsis_elevenlabs/src/sinapsis_elevenlabs/helpers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sinapsis_elevenlabs/src/sinapsis_elevenlabs/helpers/env_var_keys.py,sha256=j8J64iplBNaff1WvmfJ03eJozE1f5SdqtqQeldV2vPY,998
+sinapsis_elevenlabs/src/sinapsis_elevenlabs/helpers/voice_utils.py,sha256=fR1r1aaoFy_rQGfJLunUNdZfVxDyAo7shevS4TAXH_M,2420
+sinapsis_elevenlabs/src/sinapsis_elevenlabs/templates/__init__.py,sha256=pyTWPBLN_P6sxFTF1QqfL7iTZd9E0EaggpfwB0qLLHI,579
+sinapsis_elevenlabs/src/sinapsis_elevenlabs/templates/elevenlabs_base.py,sha256=MQglkwvyOVk4krXTXoMSPZ4yCeDBq9vMpI3riz87aIg,8291
+sinapsis_elevenlabs/src/sinapsis_elevenlabs/templates/elevenlabs_tts.py,sha256=WVTROfB2ODAksHmWwV5RKcub3Hoc29OM_eAw75c9yio,2847
+sinapsis_elevenlabs/src/sinapsis_elevenlabs/templates/elevenlabs_voice_generation.py,sha256=bKo7zhfsiZwsn-qZx_MCVAIx_MmaKnaP3lc-07AwAaY,2819
+sinapsis_f5_tts/src/sinapsis_f5_tts/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sinapsis_f5_tts/src/sinapsis_f5_tts/templates/__init__.py,sha256=28BOPAr9GG1jYcrXi45ZWO1n2FAZJOdDcmRkOXdEYmk,496
+sinapsis_f5_tts/src/sinapsis_f5_tts/templates/f5_tts_inference.py,sha256=7EBxw-tRthbPDz0zFopaLdBhv7DXwxyMGXam6F1MwGs,15802
+sinapsis_zonos/src/sinapsis_zonos/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sinapsis_zonos/src/sinapsis_zonos/helpers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sinapsis_zonos/src/sinapsis_zonos/helpers/zonos_keys.py,sha256=m1GdOYfzP73JGmtxH30mNiqbNkzFsQl9o2QaT7QxSVU,2470
+sinapsis_zonos/src/sinapsis_zonos/helpers/zonos_tts_utils.py,sha256=8Tr2YgxjBfRqv_Hf6sw36X2pLzW7fdQWqa6QPBxNZK8,6419
+sinapsis_zonos/src/sinapsis_zonos/templates/__init__.py,sha256=A-_F0K3hbEFqeWWAh4YftgU9CFX-WHrauSiCAww9yp8,482
+sinapsis_zonos/src/sinapsis_zonos/templates/zonos_tts.py,sha256=KsNuT8cFTTjTEqjfEWsIr4B-DjGhVacSw2SdPckuFvk,7507
+sinapsis_speech-0.2.0.dist-info/METADATA,sha256=-qJhZCqgMvFKr7iZBbv6lIleFa2DCTb0wXp1B2dKs18,48741
+sinapsis_speech-0.2.0.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
+sinapsis_speech-0.2.0.dist-info/top_level.txt,sha256=vQFjL84TMSRld2lKvEVMUNyY2b3AVluCT1Ijws7o7_c,51
+sinapsis_speech-0.2.0.dist-info/RECORD,,

{sinapsis_speech-0.1.0.dist-info → sinapsis_speech-0.2.0.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (77.0.1)
+Generator: setuptools (78.1.0)
 Root-Is-Purelib: true
 Tag: py3-none-any
 

sinapsis_speech-0.2.0.dist-info/top_level.txt

@@ -0,0 +1,3 @@
+sinapsis_elevenlabs
+sinapsis_f5_tts
+sinapsis_zonos

sinapsis_zonos/src/sinapsis_zonos/helpers/zonos_keys.py

@@ -0,0 +1,67 @@
+# -*- coding: utf-8 -*-
+from typing import Literal
+
+from pydantic import BaseModel
+from pydantic.dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class TTSKeys:
+    """
+    A class to hold constants for the keys used in the Text-to-Speech (TTS) model configuration.
+
+    These keys represent standard fields that are used to configure various parameters of the TTS model,
+    such as speaker attributes, emotions, and other audio-related settings. They are typically used in
+    templates and potentially a TTS web application to adjust and access specific TTS settings."
+    """
+
+    speaker: Literal["speaker"] = "speaker"
+    emotion: Literal["emotion"] = "emotion"
+    vqscore_8: Literal["vqscore_8"] = "vqscore_8"
+    fmax: Literal["fmax"] = "fmax"
+    pitch_std: Literal["pitch_std"] = "pitch_std"
+    speaking_rate: Literal["speaking_rate"] = "speaking_rate"
+    dnsmos_ovrl: Literal["dnsmos_ovrl"] = "dnsmos_ovrl"
+    speaker_noised: Literal["speaker_noised"] = "speaker_noised"
+    wav: Literal["wav"] = "wav"
+    en_language: Literal["en-us"] = "en-us"
+    min_p: Literal["min_p"] = "min_p"
+
+
+class SamplingParams(BaseModel):
+    """
+    A class to hold the sampling parameters for the TTS model.
+
+    Attributes:
+        min_p (float): Minimum token probability, scaled by the highest token probability. Range: 0-1. Default: 0.0.
+        top_k (int): Number of top tokens to sample from. Range: 0-1024. Default: 0.
+        top_p (float): Cumulative probability threshold for nucleus sampling. Range: 0-1. Default: 0.0.
+        linear (float): Controls the token unusualness. Range: -2.0 to 2.0. Default: 0.0.
+        conf (float): Confidence level for randomness. Range: -2.0 to 2.0. Default: 0.0.
+        quad (float): Controls how much low probabilities are reduced. Range: -2.0 to 2.0. Default: 0.0.
+    """
+
+    min_p: float = 0.0
+    top_k: int = 0
+    top_p: float = 0.0
+    linear: float = 0.0
+    conf: float = 0.0
+    quad: float = 0.0
+
+
+class EmotionsConfig(BaseModel):
+    """
+    A class to hold emotional attributes that influence the tone of the generated speech.
+
+    These emotions are represented as float values and are used to adjust the emotional tone of the speech.
+    Higher values can represent a stronger presence of a particular emotion.
+    """
+
+    happiness: float = 0
+    sadness: float = 0
+    disgust: float = 0
+    fear: float = 0
+    surprise: float = 0
+    anger: float = 0
+    other: float = 0
+    neutral: float = 0

sinapsis_zonos/src/sinapsis_zonos/helpers/zonos_tts_utils.py

@@ -0,0 +1,153 @@
+# -*- coding: utf-8 -*-
+from typing import Set
+
+import torch
+import torchaudio
+from sinapsis_core.template_base.template import TemplateAttributeType
+from sinapsis_core.utils.logging_utils import sinapsis_logger
+from zonos.conditioning import make_cond_dict, supported_language_codes
+from zonos.model import Zonos
+
+from sinapsis_zonos.helpers.zonos_keys import SamplingParams, TTSKeys
+
+
+def get_audio_prefix_codes(prefix_path: str | None, model: Zonos) -> torch.Tensor | None:
+    """Generates audio prefix codes from an audio file.
+
+    Args:
+        prefix_path (str): Path to the audio file to generate the prefix codes from.
+        model (Zonos): The Zonos model used to generate the audio prefix codes.
+
+    Returns:
+        torch.Tensor | None: The generated audio prefix codes if available, otherwise None.
+    """
+    if prefix_path:
+        waveform, sample_rate = torchaudio.load(prefix_path)
+        waveform = waveform.mean(0, keepdim=True)
+        waveform = model.autoencoder.preprocess(waveform, sample_rate)
+        return model.autoencoder.encode(waveform.unsqueeze(0))
+    return None
+
+
+def get_conditioning(
+    attributes: TemplateAttributeType, model: Zonos, input_text: str, device: torch.device
+) -> torch.Tensor:
+    """
+    Generates conditioning tensor for the input text, combining it with speaker embeddings and emotions.
+
+    Args:
+        attributes (TemplateAttributeType): attributes with configuration for the conditioning dictionary
+        of the model.
+        model (Zonos): Model to be used during inference, where the setup is modified.
+        input_text (str): The text to be converted to speech.
+        device (torch.device): Device where model should be loaded.
+
+    Returns:
+        torch.Tensor: The generated conditioning tensor for speech synthesis.
+    """
+    speaker_embedding = get_speaker_embedding(attributes.speaker_audio, attributes.unconditional_keys, model, device)
+    emotion_data = get_emotion_tensor(attributes, device)
+    validate_language(attributes)
+
+    vq_data = torch.tensor([attributes.vq_score] * 8, device=device).unsqueeze(0)
+
+    conditioning_dict = make_cond_dict(
+        text=input_text,
+        language=attributes.language,
+        speaker=speaker_embedding,
+        emotion=emotion_data,
+        vqscore_8=vq_data,
+        fmax=attributes.fmax,
+        pitch_std=attributes.pitch_std,
+        speaking_rate=attributes.speaking_rate,
+        dnsmos_ovrl=attributes.dnsmos,
+        speaker_noised=attributes.denoised_speaker,
+        device=device,
+        unconditional_keys=attributes.unconditional_keys,
+    )
+    return model.prepare_conditioning(conditioning_dict)
+
+
+def get_emotion_tensor(attributes: TemplateAttributeType, device: torch.device) -> torch.Tensor:
+    """
+    Extracts or constructs an emotion tensor from the given attributes.
+
+    If `attributes.emotions` is present, its values are serialized and converted into a tensor.
+    If not, a default zero tensor of shape (8,) is returned, and the `emotion` key is
+    added to `attributes.unconditional_keys` (if not already included) to indicate unconditional conditioning.
+
+    Args:
+        attributes (TemplateAttributeType): Attributes for Zonos TTS model configuration.
+        device (torch.device): The device on which the tensor should be created.
+
+    Returns:
+        torch.Tensor: A tensor representing emotion values, either user-provided or default.
+    """
+    if attributes.emotions:
+        emotion_values = list(map(float, attributes.emotions.model_dump().values()))
+        return torch.tensor(emotion_values, device=device)
+    else:
+        if TTSKeys.emotion not in attributes.unconditional_keys:
+            attributes.unconditional_keys.add(TTSKeys.emotion)
+        return torch.tensor([0.0] * 8, device=device)
+
+
+def get_sampling_params(sampling_params: SamplingParams | dict) -> dict:
+    """
+    Returns a dictionary of sampling parameters for audio generation.
+
+    If `sampling_params` is a Pydantic model, its non-null fields are serialized using `model_dump()`.
+    If `sampling_params` is empty, a default dictionary with a minimum probability value is returned.
+
+    Args:
+        sampling_params (SamplingParams | dict): A SamplingParams Pydantic model or dictionary.
+
+    Returns:
+        dict: A dictionary of sampling parameters, either user-defined or with a default fallback.
+    """
+    if isinstance(sampling_params, SamplingParams):
+        return sampling_params.model_dump(exclude_none=True)
+    return {TTSKeys.min_p: 0.1}
+
+
+def get_speaker_embedding(
+    speaker_path: str | None, unconditional_keys: Set[str], model: Zonos, device: torch.device
+) -> torch.Tensor | None:
+    """Extracts speaker embedding from an audio file.
+
+    Args:
+        speaker_path (str): Path to the audio file from which the speaker embedding will be extracted.
+        unconditional_keys (dict): Dictionary of keys to condition speech synthesis.
+            This will be used to determine whether a speaker embedding is needed.
+        model (Zonos): The Zonos model used for generating the speaker embedding.
+
+    Returns:
+        torch.Tensor | None: The speaker embedding if available, otherwise None.
+    """
+    if speaker_path and TTSKeys.speaker not in unconditional_keys:
+        waveform, sample_rate = torchaudio.load(speaker_path)
+        speaker_embedding = model.make_speaker_embedding(waveform, sample_rate)
+        return speaker_embedding.to(device, dtype=torch.bfloat16)
+    return None
+
+
+def init_seed(attributes: TemplateAttributeType) -> None:
+    """Initializes the seed for reproducible results."""
+    if attributes.randomized_seed:
+        attributes.seed = torch.randint(0, 2**32 - 1, (1,)).item()
+    torch.manual_seed(attributes.seed)
+
+
+def validate_language(attributes: TemplateAttributeType) -> None:
+    """
+    Validates and updates the language attribute in the provided TTS configuration.
+
+    Checks if `attributes.language` is included in the list of supported language codes.
+    If the language is unsupported, logs an error and defaults it to `TTSKeys.en_language`.
+
+    Args:
+        attributes (TemplateAttributeType): The model attributes containing the language setting.
+    """
+    if attributes.language not in supported_language_codes:
+        sinapsis_logger.error(f"Language {attributes.language} not supported. Defaulting to {TTSKeys.en_language}")
+        attributes.language = TTSKeys.en_language

sinapsis_zonos/src/sinapsis_zonos/templates/__init__.py

@@ -0,0 +1,20 @@
+# -*- coding: utf-8 -*-
+import importlib
+from typing import Callable
+
+_root_lib_path = "sinapsis_zonos.templates"
+
+_template_lookup = {
+    "ZonosTTS": f"{_root_lib_path}.zonos_tts",
+}
+
+
+def __getattr__(name: str) -> Callable:
+    if name in _template_lookup:
+        module = importlib.import_module(_template_lookup[name])
+        return getattr(module, name)
+
+    raise AttributeError(f"template `{name}` not found in {_root_lib_path}")
+
+
+__all__ = list(_template_lookup.keys())