test(tts): Verify Kokoro TTS implementation (#183)

[m96-chan]

Overview

Test verification for PR #185 (fix for Issue #179 - TTS beep sound bug).

Test Items

Unit Tests (Added to PR #185)

• ALBERTEncoder forward pass produces valid hidden states
• KokoroTextEncoder forward pass produces valid features
• WeightNormConv1d computes correct normalized weights
• InstanceNorm1d normalizes correctly
• AdaIN applies style conditioning correctly
• build_albert_from_weights() constructs encoder from weight dict
• build_text_encoder_from_weights() constructs text encoder from weight dict

Test file: tests/test_tts_layers.py (12 tests)

Integration Tests

• KokoroModel.from_pretrained() loads model without errors
• KokoroModel.synthesize() runs without exceptions
• Output audio has correct sample rate (24kHz)
• Output audio duration is proportional to input text length
• No 440Hz beep in output audio

End-to-End Tests

• Generate audio from "Hello, world!" text
• Generate audio with different voices (af_heart, etc.)
• Save audio to WAV file and verify playback
• Compare output with reference implementation (if available)

Test Environment

• Model: Kokoro-82M (hexgrad/Kokoro-82M)
• Voice: af_heart (default)
• Hardware: GPU with CUDA support

Related

• Fixes: bug(tts): Kokoro TTS outputs 440Hz sine wave instead of speech #179
• PR: fix(tts): Remove 440Hz beep, implement ALBERT encoder (#179) #185

Notes

Full audio generation quality depends on decoder/vocoder implementation which is still in progress. Current tests should verify the text encoding pipeline works correctly.