Merge pull request #273 from MacPaw/bug/create-speech-response-decoding

Bug/create speech response decoding

Author: nezhyborets

Demo/DemoChat/Sources/SpeechStore.swift

@@ -42,7 +42,7 @@ public final class SpeechStore: ObservableObject {
                                           format: query.responseFormat?.rawValue ?? AudioSpeechQuery.AudioSpeechResponseFormat.mp3.rawValue)
             audioObjects.append(audioObject)
         } catch {
-            print(error.localizedDescription)
+            print("[SpeechStore] createSpeech error: \(error.localizedDescription)")
         }
     }
 

README.md

@@ -98,7 +98,9 @@ let configuration = OpenAI.Configuration(token: "YOUR_TOKEN_HERE", organizationI
 let openAI = OpenAI(configuration: configuration)
 ```
 
-Once token you posses the token, and the instance is initialized you are ready to make requests.
+See `OpenAI.Configuration` for more values that can be passed on init for customization, like: `host`, `basePath`, `port`, `scheme` and `customHeaders`.
+
+Once you posses the token, and the instance is initialized you are ready to make requests.
 
 ### Chats
 

Sources/OpenAI/OpenAI+OpenAIAsync.swift

@@ -79,7 +79,7 @@ extension OpenAI: OpenAIAsync {
     }
 
     public func audioCreateSpeech(query: AudioSpeechQuery) async throws -> AudioSpeechResult {
-        try await performRequestAsync(
+        try await performSpeechRequestAsync(
             request: makeAudioCreateSpeechRequest(query: query)
         )
     }
@@ -214,4 +214,36 @@ extension OpenAI: OpenAIAsync {
             }
         }
     }
+    
+    func performSpeechRequestAsync(request: any URLRequestBuildable) async throws -> AudioSpeechResult {
+        let urlRequest = try request.build(configuration: configuration)
+        if #available(iOS 15.0, macOS 12.0, tvOS 15.0, watchOS 8.0, *) {
+            let (data, _) = try await session.data(for: urlRequest, delegate: nil)
+            return .init(audio: data)
+        } else {
+            let dataTaskStore = URLSessionDataTaskStore()
+            return try await withTaskCancellationHandler {
+                return try await withCheckedThrowingContinuation { continuation in
+                    let dataTask = self.makeRawResponseDataTask(forRequest: urlRequest) { result in
+                        switch result {
+                        case .success(let success):
+                            continuation.resume(returning: .init(audio: success))
+                        case .failure(let failure):
+                            continuation.resume(throwing: failure)
+                        }
+                    }
+                    
+                    dataTask.resume()
+                    
+                    Task {
+                        await dataTaskStore.setDataTask(dataTask)
+                    }
+                }
+            } onCancel: {
+                Task {
+                    await dataTaskStore.getDataTask()?.cancel()
+                }
+            }
+        }
+    }
 }

Sources/OpenAI/OpenAI+OpenAICombine.swift

@@ -76,7 +76,7 @@ extension OpenAI: OpenAICombine {
     }
 
     public func audioCreateSpeech(query: AudioSpeechQuery) -> AnyPublisher<AudioSpeechResult, Error> {
-        performRequestCombine(
+        performSpeechRequestCombine(
             request: makeAudioCreateSpeechRequest(query: query)
         )
     }
@@ -196,5 +196,19 @@ extension OpenAI: OpenAICombine {
                 .eraseToAnyPublisher()
         }
     }
+    
+    func performSpeechRequestCombine(request: any URLRequestBuildable) -> AnyPublisher<AudioSpeechResult, Error> {
+        do {
+            let request = try request.build(configuration: configuration)
+            return session
+                .dataTaskPublisher(for: request)
+                .tryMap { (data, response) in
+                    return .init(audio: data)
+                }.eraseToAnyPublisher()
+        } catch {
+            return Fail(outputType: AudioSpeechResult.self, failure: error)
+                .eraseToAnyPublisher()
+        }
+    }
 }
 #endif

Sources/OpenAI/OpenAI.swift

@@ -295,6 +295,19 @@ extension OpenAI {
             }
         }
     }
+    
+    func makeRawResponseDataTask(forRequest request: URLRequest, completion: @escaping (Result<Data, Error>) -> Void) -> URLSessionDataTaskProtocol {
+        session.dataTask(with: request) { data, _, error in
+            if let error = error {
+                return completion(.failure(error))
+            }
+            guard let data = data else {
+                return completion(.failure(OpenAIError.emptyData))
+            }
+            
+            completion(.success(data))
+        }
+    }
 }
 
 extension OpenAI {

Tests/OpenAITests/OpenAITests.swift

@@ -294,6 +294,14 @@ class OpenAITests: XCTestCase {
 
         XCTAssertEqual(query.speed, "\(4.0)")
     }
+    
+    func testAudioCreateSpeech() async throws {
+        let query = AudioSpeechQuery(model: .tts_1, input: "Hello, world!", voice: .alloy, speed: nil)
+        let data = Data(repeating: 11, count: 11)
+        urlSession.dataTask = .successful(with: data)
+        let response = try await openAI.audioCreateSpeech(query: query)
+        XCTAssertEqual(response.audio, data)
+    }
 
     func testAudioSpeechError() async throws {
         let query = AudioSpeechQuery(model: .tts_1, input: "Hello, world!", voice: .alloy, responseFormat: .mp3, speed: 1.0)

Tests/OpenAITests/OpenAITestsCombine.swift

@@ -75,6 +75,14 @@ final class OpenAITestsCombine: XCTestCase {
         XCTAssertEqual(result, moderationsResult)
     }
 
+    func testAudioCreateSpeech() throws {
+        let query = AudioSpeechQuery(model: .tts_1, input: "Hello, world!", voice: .alloy, speed: nil)
+        let data = Data(repeating: 10, count: 10)
+        urlSession.dataTask = .successful(with: data)
+        let response = try awaitPublisher(openAI.audioCreateSpeech(query: query), timeout: 1)
+        XCTAssertEqual(data, response.audio)
+    }
+    
     func testAudioTranscriptions() throws {
         let data = Data()
         let query = AudioTranscriptionQuery(file: data, fileType: .m4a, model: .whisper_1)