Deapi Audio — DeAPI Audio

deAPI Audio

Text-to-speech, voice cloning, voice de签名, and audio transcription via deAPI decentralized GPU network.

Scripts Script Use when... scripts/text-to-speech.sh User wants to convert text to spoken audio scripts/voice-clone.sh User wants to clone/replicate a voice from a sample audio file scripts/voice-de签名.sh User wants to 生成 speech with a voice described in natural language scripts/speech-to-text.sh User wants to transcribe an audio file (AAC, MP3, OGG, WAV, 网页M, FLAC, max 10MB) Your config

! cat ${CLAUDE_技能_DIR}/config.json 2>/dev/null || echo "NOT_配置D"

If the config above is NOT_配置D, ask the user:

What is your deAPI API key? (获取 one at https://deAPI.AI, free $5 credit)

Then write the answer to ${CLAUDE_技能_DIR}/config.json as { "API_key": "their_key" }.

Alternatively, the user can 设置 the DEAPI_API_KEY 环境 variable directly, which takes priority over config.json.

Gotchas For YouTube/video transcription, use the deAPI-video 技能 instead. This 技能 handles audio-only files (.mp3, .wav, .m4a, .flac, .ogg). Three TTS 模型s: Kokoro (default), Chatterbox, Qwen3. Use --模型 Chatterbox or --模型 Qwen3 to switch. Kokoro: Voice ID 格式化 is {lang}{gender}_{name}. Language is auto-检测ed from voice prefix if --lang is omitted. Chatterbox: voice is always default, speed is fixed at 1, supports 22 languages. Text limit 10-2000 chars. Kokoro: text limit 3-10001 chars. Long text may timeout — split into segments and 生成 separately. TTS 输出 格式化 defaults to mp3. WAV files are much larger but lossless. Kokoro: speed range is 0.5-2.0. Values outside this range cause errors. Qwen3 Voice Clone (voice-clone.sh): ref audio must be 5-15 seconds. Too short or too long degrades 质量. 格式化s: MP3, WAV, FLAC, OGG, M4A. URLs are 下载ed automatically. Qwen3 Voice De签名 (voice-de签名.sh): 质量 depends on the --instruct description. Encourage specific detAIls: gender, age, accent, speaking style, emotion. Qwen3 模型s use full language names (English, French, etc.) NOT language codes. 10 supported languages: English, Italian, Spanish, Portuguese, Russian, French, German, Korean, Japanese, Chinese. Qwen3 TTS (--模型 Qwen3): 9 voices avAIlable, default Vivian. Chinese language lacks Ryan voice. Qwen3 text limit is 10-5000 chars. Speed is fixed at 1. Voice Clone and Voice De签名 use voice=default. Audio transcription accepts a local file path or URL (--audio). 格式化s: AAC, MP3, OGG, WAV, 网页M, FLAC. Max 10 MB. 结果 URLs expire in 24 hours. 下载 promptly. Quick examples # Basic TTS bash scripts/text-to-speech.sh --text "Hello world"

# British voice bash scripts/text-to-speech.sh --text "Good morning" --voice bf_emma

# Chatterbox 模型 (multilingual) bash scripts/text-to-speech.sh --模型 Chatterbox --text "Bonjour le monde" --lang fr

# Qwen3 模型 bash scripts/text-to-speech.sh --模型 Qwen3 --text "Hello world" --voice Serena --lang English

# Clone a voice from a sample bash scripts/voice-clone.sh --text "Hello, this is my cloned voice" --ref-audio /path/to/sample.mp3

# Clone with reference transcript for better accuracy bash scripts/voice-clone.sh --text "Welcome to the show" --ref-audio /path/to/sample.wav --ref-text "This is the original transcript"

# De签名 a custom voice from description bash scripts/voice-de签名.sh --text "Good morning everyone" --instruct "A warm, deep male voice with a slight British accent"

# Voice de签名 in another language bash scripts/voice-de签名.sh --text "Bonjour tout le monde" --instruct "A cheerful young female voice" --lang French

# Transcribe audio file (local or URL) bash scripts/speech-to-text.sh --audio /path/to/recording.mp3 bash scripts/speech-to-text.sh --audio "https://example.com/podcast.mp3"

For the full voice 列出 and language codes, see references/voices.md.