fal-audio（福尔音频）

fal.ai Audio

Text-to-speech and speech-to-text using state-of-the-art audio models on fal.ai.

How It Works

1. User provides text (for TTS) or audio URL (for STT)
2. Script selects appropriate model
3. Sends request to fal.ai API
4. Returns audio URL (TTS) or transcription text (STT)

Finding Models

To discover the best and latest audio models, use the search API:

# Search for text-to-speech models
bash /mnt/skills/user/fal-generate/scripts/search-models.sh --category "text-to-speech"

# Search for speech-to-text models
bash /mnt/skills/user/fal-generate/scripts/search-models.sh --category "speech-to-text"

# Search for music generation models
bash /mnt/skills/user/fal-generate/scripts/search-models.sh --query "music generation"

Or use the search_models MCP tool with relevant keywords like "tts", "speech", "music".

Usage

Text-to-Speech

bash /mnt/skills/user/fal-audio/scripts/text-to-speech.sh [options]

Arguments:

• --text - Text to convert to speech (required)
• --model - TTS model (defaults to fal-ai/minimax/speech-2.8-turbo)
• --voice - Voice ID or name (model-specific)

Examples:

# Basic TTS (fast, good quality)
bash /mnt/skills/user/fal-audio/scripts/text-to-speech.sh \
  --text "Hello, welcome to the future of AI."

# High quality with MiniMax HD
bash /mnt/skills/user/fal-audio/scripts/text-to-speech.sh \
  --text "This is premium quality speech." \
  --model "fal-ai/minimax/speech-2.8-hd"

# Natural voices with ElevenLabs
bash /mnt/skills/user/fal-audio/scripts/text-to-speech.sh \
  --text "Natural sounding voice generation" \
  --model "fal-ai/elevenlabs/tts/eleven-v3"

# Multi-language TTS
bash /mnt/skills/user/fal-audio/scripts/text-to-speech.sh \
  --text "Bonjour, bienvenue dans le futur." \
  --model "fal-ai/chatterbox/text-to-speech/multilingual"

Speech-to-Text

bash /mnt/skills/user/fal-audio/scripts/speech-to-text.sh [options]

Arguments:

• --audio-url - URL of audio file to transcribe (required)
• --model - STT model (defaults to fal-ai/whisper)
• --language - Language code (optional, auto-detected)

Examples:

# Transcribe with Whisper
bash /mnt/skills/user/fal-audio/scripts/speech-to-text.sh \
  --audio-url "https://example.com/audio.mp3"

# Transcribe with speaker diarization
bash /mnt/skills/user/fal-audio/scripts/speech-to-text.sh \
  --audio-url "https://example.com/meeting.mp3" \
  --model "fal-ai/elevenlabs/speech-to-text/scribe-v2"

# Transcribe specific language
bash /mnt/skills/user/fal-audio/scripts/speech-to-text.sh \
  --audio-url "https://example.com/spanish.mp3" \
  --language "es"

MCP Tool Alternative

Use search_models MCP tool or search-models.sh to find the best current model, then call mcp__fal-ai__generate with the discovered modelId.

Output

Text-to-Speech Output

Generating speech...
Model: fal-ai/minimax/speech-2.8-turbo

Speech generated!

Audio URL: https://v3.fal.media/files/abc123/speech.mp3
Duration: 5.2s

Speech-to-Text Output

Transcribing audio...
Model: fal-ai/whisper

Transcription complete!

Text: "Hello, this is the transcribed text from the audio file."
Duration: 12.5s
Language: en

Present Results to User

For TTS:

Here's the generated speech:

[Download audio](https://v3.fal.media/files/.../speech.mp3)

• Duration: 5.2s | Model: Maya TTS

For STT:

Here's the transcription:

"Hello, this is the transcribed text from the audio file."

• Duration: 12.5s | Language: English

Model Selection Tips

• Text-to-Speech: Search for text-to-speech category. Consider quality vs speed tradeoffs.
• Text-to-Music: Search for music generation. Some models specialize in vocals, others in instrumental.
• Speech-to-Text: Search for speech-to-text category. Consider whether you need speaker diarization or multi-language support.

Troubleshooting

Empty Audio

Error: Generated audio is empty

Check that your text is not empty and contains valid content.

Unsupported Audio Format

Error: Audio format not supported

Supported formats: MP3, WAV, M4A, FLAC, OGG
Convert your audio to a supported format.

Language Detection Failed

Warning: Could not detect language, defaulting to English

Specify the language explicitly with --language option.