Weird "breath" synthesis at beginning of generation if you start with multiple newlines

[mashdragon]

Here's an example:

import wave
from piper.voice import PiperVoice as piper

# Load the PiperVoice model
piper_model = piper.load("en_US-amy-medium.onnx", "en_US-amy-medium.onnx.json", use_cuda=False)

# Open a WAV file for writing
wav_file = wave.open("output.wav", "wb")
wav_file.setnchannels(1)  # Mono
wav_file.setsampwidth(2)  # 16-bit
wav_file.setframerate(16000)  # 16 kHz

# Generate a text-to-speech clip using piper.voice
piper_model.synthesize("\n\nLet's begin.", wav_file)

# Close the WAV file
wav_file.close()

Here's the output (converted to mp4 for github):
https://github.com/user-attachments/assets/a3fc8f1c-05f5-4ee4-976b-04738c759259

The solution is just to strip the text you are feeding to the model. I'm posting this if anyone else has the same problem.