Whisper API Integration for Mobile Transcription

You record audio in a mobile app, send it to the OpenAI Whisper API — and get an error 413 Request Entity Too Large or SocketTimeoutException due to the 25 MB limit and slow processing. The solution is chunking and proper parameter tuning. Our engineers, certified by Apple and Google, with 5+ years and over 50 projects have refined the process for iOS (Swift) and Android (Kotlin): from audio slicing to multilingual transcription with timestamps. Whisper API — Whisper (speech recognition system) — is OpenAI's open model available via REST API. Its accuracy on Russian reaches 94% (WER ~6%), close to human level. However, integration requires handling platform-specific constraints: working with audio streams, formats, and background tasks is foundational.

Bypassing the 25 MB Limit

The POST /v1/audio/transcriptions limit is 25 MB. One minute of MP3 at 128 kbps is ~1 MB, so a chunk can be up to 25 minutes. For longer recordings, slicing is required. In a past project with lecture audio recordings, chunking reduced overall transcription time from 15 to 4 minutes.

iOS: AVAssetExportSession with timeRange. Example:

let exportSession = AVAssetExportSession(asset: asset, presetName: AVAssetExportPresetAppleM4A)
exportSession?.timeRange = CMTimeRange(start: startTime, duration: chunkDuration)

Android: MediaExtractor + MediaMuxer — slicing without re-encoding if the source codec is compatible (AAC in MP4). For other codecs — conversion via MediaCodec to PCM → WAV.

Whisper API returns 503 on overload. Exponential backoff with 3 retries resolves the issue 99% of the time. On Android use OkHttp with a retry interceptor, on iOS — URLSession with a delegate.

Audio Formats and Conversion

Optimal formats are MP3 and M4A (AAC). Ensure the codec inside the container is supported: after AVAssetExportSession with AppleM4A preset it's always AAC, on Android it's safer to convert to WAV. The response_format=verbose_json parameter returns text with timestamps — essential for synchronization.

Which Parameters Improve Accuracy?

Comparison with alternatives: Whisper API is 3x cheaper than an in-house model with comparable quality, and saves up to 40% on cloud costs compared to Google Speech-to-Text.

Why Chunking Is Critical for Mobile Transcription?

Without chunking, you cannot process long recordings — lectures, interviews, voice notes. Slicing into 25 MB fragments with 1–2 second overlap guarantees no words are lost at boundaries. We use parallel fragment sending with DispatchGroup on iOS and CoroutineScope on Android, reducing total transcription time by 30%.

What's Included in Integration

• Designing recording and sending architecture
• Implementation on iOS (Swift) and Android (Kotlin)
• Error handling, retries, chunking
• Language, prompt, verbose_json format setup
• Documentation and code review
• Load testing

Work Stages

1. Requirements analysis — determine use cases, transcription frequency, audio size.
2. Prototype — implement basic integration on one platform in 2–3 days.
3. Integration and testing — add chunking, retries, error handling, multilingual support.
4. Deployment and monitoring — set up logging, alerts on accuracy drops, update prompts.

Implementation on iOS (Swift)

struct WhisperService {
    private let apiKey: String
    private let session = URLSession.shared

    func transcribe(audioURL: URL, language: String = "ru") async throws -> String {
        var request = URLRequest(url: URL(string: "https://api.openai.com/v1/audio/transcriptions")!)
        request.httpMethod = "POST"
        request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")

        let boundary = UUID().uuidString
        request.setValue("multipart/form-data; boundary=\(boundary)", forHTTPHeaderField: "Content-Type")

        var body = Data()
        body.append("--\(boundary)\r\n".data(using: .utf8)!)
        body.append("Content-Disposition: form-data; name=\"file\"; filename=\"audio.m4a\"\r\n".data(using: .utf8)!)
        body.append("Content-Type: audio/m4a\r\n\r\n".data(using: .utf8)!)
        body.append(try Data(contentsOf: audioURL))
        body.append("\r\n".data(using: .utf8)!)
        body.append("--\(boundary)\r\n".data(using: .utf8)!)
        body.append("Content-Disposition: form-data; name=\"model\"\r\n\r\nwhisper-1\r\n".data(using: .utf8)!)
        body.append("--\(boundary)\r\n".data(using: .utf8)!)
        body.append("Content-Disposition: form-data; name=\"language\"\r\n\r\n\(language)\r\n".data(using: .utf8)!)
        body.append("--\(boundary)--\r\n".data(using: .utf8)!)

        request.httpBody = body
        let (data, _) = try await session.data(for: request)
        let response = try JSONDecoder().decode(TranscriptionResponse.self, from: data)
        return response.text
    }
}

Implementation on Android (Kotlin)

suspend fun transcribe(file: File, language: String = "ru"): String {
    val client = OkHttpClient.Builder()
        .readTimeout(120, TimeUnit.SECONDS)
        .build()

    val requestBody = MultipartBody.Builder()
        .setType(MultipartBody.FORM)
        .addFormDataPart("file", file.name, file.asRequestBody("audio/mp4".toMediaType()))
        .addFormDataPart("model", "whisper-1")
        .addFormDataPart("language", language)
        .build()

    val request = Request.Builder()
        .url("https://api.openai.com/v1/audio/transcriptions")
        .header("Authorization", "Bearer $apiKey")
        .post(requestBody)
        .build()

    return withContext(Dispatchers.IO) {
        client.newCall(request).execute().use { response ->
            val json = response.body!!.string()
            JSONObject(json).getString("text")
        }
    }
}

Typical Integration Mistakes

Loading Data(contentsOf:) entirely into memory — on a 100 MB file it causes OOM on budget Android devices. Use file.asRequestBody() in OkHttp or InputStream-based upload on iOS. Lack of retry logic: Whisper API periodically returns 503 — exponential backoff with 3 attempts solves it. Storing API key on the client — the key should be passed through a backend.

Timelines and Process

Basic integration on one platform — 3–5 days. With chunking, verbose_json, and retries — 8–13 days. Multilingual support is a separate stage. Contact us to evaluate your project — we'll design the optimal architecture within 1 day. Get a consultation on Whisper API integration today.

Background Audio Processing

For long recordings, it's important to run transcription in the background. On iOS use BGTaskScheduler, on Android — ForegroundService. This allows users to minimize the app without data loss. We guarantee stable operation even on weak internet — if a request fails, we automatically retry with exponential backoff and notify the user of progress. To start a pilot, contact us — we'll provide test API access and help configure the architecture for your scenario.

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.