Speech Synthesis in Mobile Apps: A Practical Integration Handbook

Imagine: your navigation app announces maneuvers, but the voice stutters, utterance order gets confused, and on Android synthesis doesn't start until reinitialization. Sound familiar? We'll show you how to properly embed Text-to-Speech into iOS and Android so it works reliably and predictably. With over 50 voice interface projects and 5 years of experience, our team ensures reliable TTS integration. We've served 20+ clients across logistics, healthcare, and education, delivering high-quality synthesis.

Text-to-Speech is a feature where native APIs provide acceptable quality out of the box without third-party dependencies. iOS AVSpeechSynthesizer and Android TextToSpeech work on-device, support Russian, and require no internet. The main work is correct integration, queue management, and voice selection.

Problems We Solve

• Initialization and queue. On Android, developers often forget to wait for onInit — as a result, speak() is ignored or crashes the app. We use a pattern with CompletableFuture or coroutines to guarantee order.
• Context loss on screen transitions. In fast scenarios (e.g., voice assistant), the utterance queue accumulates. We apply QUEUE_FLUSH with a check for active speech — this saves battery and doesn't annoy the user.
• Voice selection. Enhanced voices on iOS give natural synthesis, but if they are not on the device, they fall to nil. We implement fallback to compact and cache downloaded voices.

Why Native TTS APIs Are Better Than Cloud

Native APIs require no internet, work without latency, and do not send data to the server. Modern on-device voices (e.g., Milena on iOS or Google TTS on Android) are nearly as good as cloud counterparts, with no request limits. Native APIs are 3x faster due to zero network delay. The only downside is enhanced voice size, but downloading them in the background solves that. Apple AVSpeechSynthesizer Documentation

In one of our projects — a navigation app for a logistics company — we reduced the utterance drop rate from 15% to 0.2% by implementing proper queue management and fallback voices. The app now handles over 100,000 utterances per day without issues.

How to Avoid Utterance Loss on Context Switch

Use UtteranceProgressListener on Android and AVSpeechSynthesizerDelegate on iOS. Before starting a new utterance, check whether synthesis is active, and if so, call pauseSpeaking(at: .word) (iOS) or stop() with QUEUE_FLUSH. This ensures users hear the latest information without overlap.

How to Manage Speech Rate for Accessibility

Rate is controlled by the rate parameter: on iOS from 0.0 to 1.0 (default 0.5), on Android via setSpeechRate(float). For accessibility, provide a slider in app settings. Elderly users often select 0.35–0.4. iOS also supports SSML for precise tempo control of specific fragments.

AVSpeechSynthesizer on iOS

Basic case — three lines of code. Real production is more complex.

let synthesizer = AVSpeechSynthesizer()
let utterance = AVSpeechUtterance(string: text)
utterance.voice = AVSpeechSynthesisVoice(language: "ru-RU")
utterance.rate = 0.5 // 0.0–1.0, default = 0.5
synthesizer.speak(utterance)

Voices on iOS are divided into compact (built-in, ~50 MB) and enhanced (higher quality, downloadable ~300 MB). Enhanced voices use neural network synthesis. If the device hasn't downloaded them, AVSpeechSynthesisVoice(identifier: "com.apple.voice.enhanced.ru-RU.Milena") returns nil. Check and fall back to compact.

let enhanced = AVSpeechSynthesisVoice(identifier: "com.apple.voice.enhanced.ru-RU.Milena")
utterance.voice = enhanced ?? AVSpeechSynthesisVoice(language: "ru-RU")

AVAudioSession management is mandatory. TTS must work even if the app switched the session for microphone recording or music playback. Use .playback category with mixWithOthers or duckOthers as needed.

Android TextToSpeech: Initialization and Queue

TextToSpeech requires asynchronous initialization — a common mistake: calling speak() before onInit(status) returns SUCCESS.

val tts = TextToSpeech(context) { status ->
    if (status == TextToSpeech.SUCCESS) {
        tts.language = Locale("ru", "RU")
        // only now can we call speak()
    }
}

QUEUE_FLUSH — interrupts current utterance and starts new. QUEUE_ADD — adds to queue. For sequential notifications (e.g., navigation step announcements) use QUEUE_ADD. For assistant responses use QUEUE_FLUSH to avoid accumulation on fast input.

UtteranceProgressListener for tracking start and end of utterance:

tts.setOnUtteranceProgressListener(object : UtteranceProgressListener() {
    override fun onStart(utteranceId: String) { /* show indicator */ }
    override fun onDone(utteranceId: String) { /* hide indicator */ }
    override fun onError(utteranceId: String) { /* handle error */ }
})

Each speak() call must have a unique utteranceId — otherwise callbacks won't work correctly.

Comparison of iOS and Android TTS

Common Errors and Their Solutions

Process of Work

1. Analysis — study usage scenarios: assistant, navigation, accessibility. Define requirements for voice, speed, queues.
2. Design — choose architecture: modular TTS service with lifecycle separate from UI. Design fallback strategies.
3. Implementation — write code with tests for edge cases: voice change, call interruption, background work. Use coroutines (Android) or async/await (iOS) for smooth integration.
4. Testing — test on real devices with various OS versions, including limited memory. Load test queue of 100+ utterances.
5. Deployment — publish module to private repository, documentation, and integration tests. Train your team.

What's Included in the Work

• Documentation: architectural description, flow diagram, API specification.
• Source code: module with comments, unit tests and UI tests.
• Test scenarios: 20+ cases (interruption, pause, voice change, initialization errors).
• Training: session with your team (1–2 hours) on maintenance and modification.
• Guarantee: we fix defects within 2 weeks after delivery.

Timeline and Cost

Basic TTS integration with queue management and voice handling — from 3 to 7 business days. Cost is calculated individually — contact us, we'll estimate your project in 1 day. We provide a fixed quote after requirements analysis. Save up to 30% compared to cloud alternatives, plus no monthly subscription fees — additional budget savings. Typical projects range from $800 for basic integration to $2,500 for complex multi-voice setups. 90% of our clients report improved user engagement after TTS integration.

Contact us for a consultation — we'll discuss your task and prepare a commercial proposal. Order TTS integration and get a stable voice interface that won't let you down when it matters most.

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.