Transform audio

Use transformAudio when the provider takes existing media and returns changed audio.

import Foundation
import SwiftAISDK

@main
struct TransformAudioExample {
    static func main() async {
        do {
            let provider = try AIProviders.elevenLabs()
            let speech = try provider.speechModel("eleven_multilingual_v2")
            let source = try await AI.generateSpeech(
                "Swift AI voice tools smoke test. This source clip is long enough for isolation.",
                using: speech,
                voice: "21m00Tcm4TlvDq8ikWAM",
                format: "mp3_32",
                providerOptions: ["elevenlabs": ["languageCode": "en"]]
            )

            let changer = try provider.voiceChangerModel()
            let changed = try await AI.transformAudio(
                audio: source.audio,
                using: changer,
                fileName: "voice.mp3",
                mimeType: source.contentType ?? "audio/mpeg",
                voice: "21m00Tcm4TlvDq8ikWAM",
                format: "mp3_32"
            )
            print("Changed voice bytes: \(changed.audio.count)")

            let isolator = try provider.voiceIsolatorModel()
            let isolated = try await AI.transformAudio(
                audio: source.audio,
                using: isolator,
                fileName: "voice.mp3",
                mimeType: source.contentType ?? "audio/mpeg",
                providerOptions: ["elevenlabs": ["fileFormat": "other"]]
            )
            print("Isolated voice bytes: \(isolated.audio.count)")
        } catch {
            print("Audio transformation failed: \(error)")
        }
    }
}

AudioTransformationRequest carries the input audio bytes, file name, MIME type, optional target voice, output format, provider options, extra body, headers, and abort signal.

AudioTransformationResult returns transformed audio bytes plus content type and metadata.

ElevenLabs exposes provider.voiceChangerModel() for /v1/speech-to-speech/:voice_id and provider.voiceIsolatorModel() for /v1/audio-isolation.