ai-agents-speech-quality-varia-during-multi-speaker-inference-output-sometimes-sw-1ae6f136

THEN Run inference multiple times with the same input text and select the best result. Use `chat.sample_random_speaker()` to generate a fixed speaker embedding for consistent timbre across runs. For critical applications, implement a loop that stores outputs and picks the one with highest confidence or lowest distortion.

```python
# Paraphrase: Retry inference to pick best output
best_wav = None
for _ in range(3):
    wav = chat.infer(["Your text"])[0]
    if best_wav is None or some_quality_check(wav):
        best_wav = wav
```