#tts

x

The Thai Voice Content Workflow, From Script to Finished Audio File By Nokka | September...

TTS ไทย 3 ตัวเทียบกัน, ThonburianTTS vs OmniVoice vs ElevenLabs...

ทำคอนเทนต์เสียงไทยด้วย AI, ขั้นตอนจริงตั้งแต่เขียนสคริปต์จนได้ไฟล์เสียง โดย Nokka (นก-กา)...

We ran three text-to-speech tools on the same job over 90 days and 879 calls. Here are the cost, speed and failure figures for each, and which one is still running today.

The word was qara, "black". I fed it to an off-the-shelf TTS model, hit play, and got back a soft,...

The narration audio had slightly weird pitch accent. Proper nouns, place names, years. "Pompeii",...

Why storing content and tts_friendly_content as two independently editable columns turned a simple straightforward into a synchronization nightmare and how I am replacing sanitization with sending style instrcutions to Qwen3-TTS prompt instead.

A series of 18 articles summarizing the failures encountered in designing voices from single-line captions, automating training corpus creation, and mass-producing 12 role-specific voices. Organized into four chapters—Design, Manufacturing, Inspection, and Operation—presented in sequential reading order.

OmniVoice โมเดล TTS 600+ ภาษา โคลนเสียงจากคลิป 3 วินาที เปิดซอร์สฟรี ทีม k2-fsa เปิดตัว...

When inspecting the quality of a speech synthesis model using only STT, we missed all added sounds lasting just 0.1 seconds after silence. I’ll explain how we built a detector using waveform envelopes and how all 12 models produced false positives due to commas.

A simple change in language code case allows Japanese to pass through English G2P, resulting in a model that completes training with empty phonemes and only speaks gibberish. This catalog also covers other pitfalls encountered during speech model training, including curl OOM errors, CUDA asserts in continuous learning, and reference audio length limits.