Voice Synthesis in Mobile Apps: Configuring OpenAI TTS

Voice Synthesis in Mobile Apps: Configuring OpenAI TTS

Voice synthesis (text-to-speech) — a technology that converts text into speech. A mobile app needs to voice long texts—news, audiobooks, voice prompts. Without optimization, users wait 3–5 seconds before playback starts. OpenAI TTS solves this, but only with proper streaming and caching. We’ll show you how to achieve sub-second latency on iOS and Android.

In this article we break down the integration architecture: from a simple REST request to streaming playback with ExoPlayer and AVAudioPlayer. We show how to cache synthesized audio and handle texts longer than 4096 characters. The result is a ready-to-deploy solution that can be integrated in 3–10 days.

What Problems Do We Solve?

Three main challenges arise: high latency, cost, and length limits. First, without streaming, you must wait for the entire file to load. Second, re-synthesizing identical text wastes API credits. Third, OpenAI TTS accepts up to 4096 characters per request, so long texts require splitting. Our solution tackles each through caching, streaming, and sentence-based splitting.

How the OpenAI TTS API Works

POST https://api.openai.com/v1/audio/speech
Authorization: Bearer {api_key}
Content-Type: application/json

{
  "model": "tts-1-hd",
  "input": "Your text here",
  "voice": "nova",
  "response_format": "mp3",
  "speed": 1.0
}

According to OpenAI documentation, two models are available. tts-1 is faster, slightly lower quality, cheaper ($15/million characters). tts-1-hd is higher quality, about 30% slower, more expensive ($30/million characters). Voices: alloy (neutral), echo (male soft), fable (British), onyx (male deep), nova (female lively), shimmer (female calm). For Russian, nova and shimmer sound most natural. The speed parameter ranges from 0.25 to 4.0, default is 1.0; values above 1.3 start to break prosody.

Why Caching Is Critical for UX

Every API call takes time and costs money. Caching avoids re-synthesizing the same text. For UI strings (greetings, hints), we pre-generate audio on first launch and cache it permanently. Estimated savings: with active caching, API costs can be reduced up to 40%, which translates to saving $200–$500 per month for active apps. Caching cuts API calls by up to 80% for repeated phrases, making it 5 times more cost-effective.

// iOS: cached synthesized audio
class TTSCache {
    private let cacheURL: URL

    init() {
        cacheURL = FileManager.default.urls(for: .cachesDirectory, in: .userDomainMask)[0]
            .appendingPathComponent("tts_cache")
        try? FileManager.default.createDirectory(at: cacheURL, withIntermediateDirectories: true)
    }

    func key(text: String, voice: String) -> String {
        let input = "\(text)|\(voice)"
        return SHA256.hash(data: Data(input.utf8)).hexString
    }

    func get(_ key: String) -> Data? {
        let url = cacheURL.appendingPathComponent(key + ".mp3")
        return try? Data(contentsOf: url)
    }

    func set(_ key: String, data: Data) {
        let url = cacheURL.appendingPathComponent(key + ".mp3")
        try? data.write(to: url)
    }
}

Before each TTS request, check the cache. A cache hit means instant playback.

Non-Streaming Implementation (for Short Texts)

// iOS: load and play
func speak(text: String, voice: String = "nova") async throws {
    var request = URLRequest(url: URL(string: "https://api.openai.com/v1/audio/speech")!)
    request.httpMethod = "POST"
    request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
    request.setValue("application/json", forHTTPHeaderField: "Content-Type")

    let body = TTSSpeechRequest(model: "tts-1", input: text, voice: voice, responseFormat: "mp3")
    request.httpBody = try JSONEncoder().encode(body)

    let (data, _) = try await URLSession.shared.data(for: request)
    audioPlayer = try AVAudioPlayer(data: data)
    audioPlayer?.play()
}

For short phrases (under 100 characters) on tts-1, latency is ~300–500 ms—acceptable without streaming. For long texts, streaming is needed.

class OpenAITTSStreamer(private val apiKey: String, private val context: Context) {
    private val exoPlayer = ExoPlayer.Builder(context).build()

    fun speak(text: String, voice: String = "nova") {
        val requestBody = JSONObject().apply {
            put("model", "tts-1")
            put("input", text)
            put("voice", voice)
            put("response_format", "mp3")
        }.toString().toRequestBody("application/json".toMediaType())

        // Use OkHttp as DataSource via a custom MediaSource
        val call = OkHttpClient().newCall(
            Request.Builder()
                .url("https://api.openai.com/v1/audio/speech")
                .header("Authorization", "Bearer $apiKey")
                .post(requestBody)
                .build()
        )

        call.enqueue(object : Callback {
            override fun onResponse(call: Call, response: Response) {
                // Write stream to temporary file, start playback simultaneously
                val tempFile = File(context.cacheDir, "tts_${System.currentTimeMillis()}.mp3")
                response.body!!.byteStream().use { input ->
                    tempFile.outputStream().use { output ->
                        val buffer = ByteArray(8192)
                        var bytes: Int
                        var firstChunk = true
                        while (input.read(buffer).also { bytes = it } != -1) {
                            output.write(buffer, 0, bytes)
                            if (firstChunk && tempFile.length() > 32768) {
                                firstChunk = false
                                // Start playback after first 32 KB
                                Handler(Looper.getMainLooper()).post {
                                    exoPlayer.setMediaItem(MediaItem.fromUri(tempFile.toUri()))
                                    exoPlayer.prepare()
                                    exoPlayer.play()
                                }
                            }
                        }
                    }
                }
            }
            override fun onFailure(call: Call, e: IOException) { /* handle error */ }
        })
    }
}

How to Handle Long Texts?

OpenAI TTS accepts up to 4096 characters per request. For long texts—split by sentences:

func splitBySentences(_ text: String, maxLength: Int = 1000) -> [String] {
    var chunks: [String] = []
    var current = ""
    for sentence in text.components(separatedBy: CharacterSet(charactersIn: ".!?\n")) {
        let trimmed = sentence.trimmingCharacters(in: .whitespaces)
        if trimmed.isEmpty { continue }
        if current.count + trimmed.count > maxLength {
            if !current.isEmpty { chunks.append(current) }
            current = trimmed
        } else {
            current += (current.isEmpty ? "" : ". ") + trimmed
        }
    }
    if !current.isEmpty { chunks.append(current) }
    return chunks
}

Synthesize chunks in parallel using TaskGroup, play sequentially—the overall latency is lower than sequential processing.

What's Included (Deliverables)

• Requirements analysis and architecture design
• Implementation of REST and streaming requests to OpenAI TTS
• Device-side caching (LRU cache with hashing)
• Long text handling (sentence-based splitting)
• Player integration (AVAudioPlayer / ExoPlayer) with background playback support
• Compliance with App Store Review Guidelines (Section 4.2) and AVAudioSession setup for iOS
• Testing on real devices (iOS 15+ / Android 10+)
• Operations documentation and API key setup guide
• Code repository access with full commit history
• Team training session (up to 2 hours) via video call
• 2 weeks of post-delivery support

Work Process

1. Analysis — Study your app, identify TTS call points, measure current latencies.
2. Prototyping — Build an MVP on a single screen, demonstrate streaming with cache.
3. Integration — Embed the ready-made modules into your codebase.
4. Testing — Load test, fine-tune for your use cases.
5. Deployment — Publish to App Store / Google Play, monitor.

Timelines and Cost

Basic integration (REST + cache) — 3–4 days ($1,500). Extended integration (streaming + long text handling + UI) — 7–10 days ($3,500). You can reduce API costs by up to 40% with caching, saving an estimated $200–$500 per month for active apps. Exact cost is calculated after auditing your project—reach out to us for an estimate. For a precise quote, contact us.

Our Team’s Experience

We have over 5 years of mobile development experience. We’ve completed 15+ projects integrating AI services, including OpenAI, Google Cloud Speech, and Yandex SpeechKit. Our developers are Apple and Google certified. We guarantee stable operation and transparent collaboration.

Get a consultation on integrating voice synthesis into your app—contact us.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.