Feature Request: Fal audio, speech, and music generation support

[tombeckenham]

Feature Request

Description

Add audio, speech, and music generation adapters to the @tanstack/ai-fal package. The fal adapter currently supports image and video generation, but fal's platform also offers 600+ models including audio modalities:

• Text-to-Speech (e.g., fal-ai/kokoro - multi-language TTS)
• Text-to-Music (e.g., fal-ai/diffrhythm - music generation from prompts/lyrics)
• Text-to-Sound Effects (e.g., sound effect generation from descriptions)
• Speech-to-Text (e.g., fal-ai/whisper, fal-ai/wizper - transcription)

Motivation

TanStack AI's fal adapter (@tanstack/ai-fal) currently implements falImage and falVideo adapters following the tree-shakeable adapter pattern. Adding audio/speech/music adapters would complete fal's media generation coverage and align with TanStack AI's goal of being a comprehensive, provider-agnostic AI SDK.

fal's audio models support:

• Multi-language text-to-speech with multiple voices
• Music generation from text prompts, lyrics, and reference audio
• Sound effect generation from descriptions
• Speech transcription and translation

Proposed API

Following the existing adapter pattern:

import { falAudio, falSpeech, falMusic } from '@tanstack/ai-fal/adapters'

// Text-to-Speech
const speechAdapter = falSpeech('fal-ai/kokoro')

// Text-to-Music  
const musicAdapter = falMusic('fal-ai/diffrhythm')

// Text-to-Sound Effects
const soundAdapter = falAudio('fal-ai/sound-effects')

Additional Context

• Existing adapters use fal.subscribe() (image) and fal.queue (video) patterns
• Audio generation may use either pattern depending on model latency
• The fal SDK (@fal-ai/client) already supports audio responses with File/Audio output types
• Model metadata types in model-meta.ts would need to be extended for audio models