Running two TTS architectures on the same small fine-tune corpus surfaces a real trade-off: F5-TTS commits hard to accent character at the cost of phonetic stability; StyleTTS2 stays phonetically stable at the cost of accent commitment. Neither dominates. Each has its own late-epoch failure mode, just different ones.
This is a write-up of the comparison, with the concrete failure modes that made the trade-off visible.
Notes on the iteration loop we ran while fine-tuning F5-TTS and StyleTTS2 on a small Northern English corpus. The headline finding is that the listening test isn't optional polish at the end — it's the only measurement that catches the failure modes that matter, and each round of listening produces specific phonetic observations that map to specific engineering decisions.
This is a write-up of the methodology, with the concrete examples that forced each decision.
Notes from getting F5-TTS, StyleTTS2, kokoro/Misaki, and whisper.cpp to work on an AMD Phenom II X6 1090T (2010 K10/Family-10h architecture).
The CPU has SSE/SSE2/SSE3/SSE4a, plus CX16/POPCNT/LAHF — but no SSE4.1, no SSE4.2, no AVX, no AVX2, no FMA, no F16C. That puts it below the modern x86-64-v2 baseline. A growing share of binary Python wheels in the AI ecosystem assume v2 or v3, so they SIGILL or SIGFPE at import. This is a ground-truth list of what we hit and what worked.
A ~250-line trainer for F5-TTS that bypasses the HuggingFace datasets and
accelerate dependency stack. Useful when:
f5-tts_finetune-cli won't install/run because of pyarrow /
pandas / datasets issues (binary-wheel CPU baseline mismatches, missing
dependencies, etc.)