Add longform narration

README.md

@@ -14,6 +14,7 @@ A lightweight Python library for running TTS models with a unified API.
 - 🎯 Focus on ease of use - a single API for all models
 - 📦 Minimal dependencies - one package for all models
 - 🔌 Extensible architecture - easily add new models
+- 💎 Feature-rich - includes longform narration, voice cloning support, and more
 
 ## Models
 
@@ -33,11 +34,13 @@ pip install simpletts
 ## Quick Start
 
 ```python
-from simpletts.models.xtts import XTTS
 import soundfile as sf
+from simpletts.models.xtts import XTTS
 
 tts = XTTS(device="auto")
-# Note: XTTS is licensed under the CPML license which restricts commercial use.
+# Easily swap out for F5-TTS:
+# from simpletts.models.f5 import F5TTS
+# tts = F5TTS(device="auto")
 
 array, sr = tts.synthesize("Hello, world!", ref="sample.wav")
 

docs/longform.md

@@ -0,0 +1,24 @@
+# Longform Narration
+
+SimpleTTS supports longform narration as an **experimental feature**. Longform narration is powered by [`txtsplit`](https://github.com/fakerybakery/txtsplit). This means it may not always work as expected. Quality may vary.
+
+## Example
+
+Here is an example of how to use the `longform` method to with Kokoro:
+
+```python
+from simpletts.models.kokoro import Kokoro
+import soundfile as sf
+
+# Initialize Kokoro model
+tts = Kokoro(device="auto")
+
+# Synthesize speech
+text = """
+Enter your longform text here...
+"""
+audio, sr = tts.longform(text, ref="af")
+
+# Save output audio
+sf.write("output.wav", audio, sr)
+```

sample/longform.py

@@ -0,0 +1,24 @@
+from simpletts.models.kokoro import Kokoro
+import soundfile as sf
+
+# Initialize Kokoro model
+tts = Kokoro(device="auto")
+
+# Synthesize speech
+text = """
+Text-to-speech technology has come a long way in recent years, with many powerful models now available to developers. However, the fragmented ecosystem of TTS libraries poses significant challenges. Each model typically comes with its own unique API, dependencies, and setup requirements, making it difficult for developers to experiment with different models or switch between them as needed.
+
+This is where a unified TTS library becomes invaluable. By providing a consistent interface across multiple models, it dramatically simplifies the development process. Developers can focus on their applications rather than wrestling with different APIs and dependencies for each model they want to try.
+
+A unified library also promotes better code maintainability and portability. When your application's TTS functionality is abstracted behind a common interface, switching models becomes as simple as changing a single line of code. This flexibility is especially important as the field of TTS continues to evolve rapidly, with new and improved models being released regularly.
+
+Additionally, a unified library can handle common tasks like text preprocessing, audio post-processing, and long-form text synthesis consistently across all models. This reduces duplication of effort and helps ensure consistent behavior regardless of the underlying model being used.
+
+For organizations, having a unified TTS library means reduced training time for developers, simplified maintenance, and the ability to easily benchmark different models against each other. It also makes it easier to swap out models based on specific needs - whether that's quality, speed, licensing requirements, or language support.
+
+In conclusion, as TTS technology becomes increasingly important in modern applications, having a unified library isn't just convenient - it's becoming essential for efficient development and maintenance of TTS-enabled applications.
+"""
+audio, sr = tts.longform(text, ref="af")
+
+# Save output audio
+sf.write("output.wav", audio, sr)

setup.py

@@ -31,6 +31,7 @@
         "tqdm",
         "openphonemizer",
         "click",
+        "txtsplit",
     ],
     extras_require={
         "xtts": [

simpletts/models/__init__.py

@@ -1,6 +1,8 @@
 from abc import ABC, abstractmethod
 import numpy as np
 from typing import Tuple, Optional, Union
+from txtsplit import txtsplit
+from tqdm import tqdm
 
 
 class TTSModel(ABC):
@@ -22,6 +24,35 @@ def synthesize(self, text: str, **kwargs) -> Tuple[np.ndarray, int]:
         """
         pass
 
+    def longform(self, text: str, **kwargs) -> Tuple[np.ndarray, int]:
+        """
+        Synthesize long text by splitting into chunks and concatenating results.
+
+        Args:
+            text: The text to synthesize
+            **kwargs: Additional arguments passed to synthesize()
+
+        Returns:
+            Tuple containing:
+                - Concatenated audio array as numpy array
+                - Sample rate as integer
+        """
+        chunks = txtsplit(text)
+
+        # Synthesize each chunk
+        audio_chunks = []
+        sr = None
+        for chunk in tqdm(chunks):
+            audio, chunk_sr = self.synthesize(chunk, **kwargs)
+            if sr is None:
+                sr = chunk_sr
+            elif sr != chunk_sr:
+                raise ValueError("Inconsistent sample rates between chunks")
+            audio_chunks.append(audio)
+
+        # Concatenate audio chunks
+        return np.concatenate(audio_chunks), sr
+
     @abstractmethod
     def __init__(self, device: Optional[str] = "auto", **kwargs):
         """