Arm/whisper-medium-int8-litert

Whisper Medium optimized for Arm-based mobile CPUs with SME2

Whisper Medium automatic speech recognition optimized as an INT8 LiteRT model for Arm-based mobile CPUs with SME2.

Summary

This repository contains an Arm-optimized version of openai/whisper-medium for automatic speech recognition. The model is provided in LiteRT (.tflite), targeting Mobile CPU systems.

This version is intended to demonstrate efficient inference on Arm-based platforms while preserving the original model's intended behavior. Arm has evaluated this model on LibriSpeech ASR and measured performance on a representative evaluation target.

Key results

Area	Result
Model format	LiteRT (.tflite)
Target device class	Mobile CPU
Reference device	vivo X300 (C1-Ultra, C1-Premium, C1-Pro, Android 16 / OriginOS 6)
Primary performance result	8853.61 ms p50 end-to-end latency (RTFx 1.45x real-time)
Accuracy result	WER 2.86%, CER 1.26%
Size / memory result	743.15 MB, 3.92x smaller than FP32

Original model

Field	Value
Original model	openai/whisper-medium
Original source	Hugging Face
Original developer	OpenAI
Original model card	openai/whisper-medium
Original license	Apache-2.0

Model files

File	Description
whisper-medium-int8-litert.tflite	Arm-optimized model for deployment
example.py	Minimal inference example
pyproject.toml	Pinned runtime dependencies for example.py, resolved with uv
uv.lock	Locked dependency resolution for pyproject.toml
config.yaml	Model I/O contract used by the example
benchmarks/	FP32 baseline and Arm-optimized benchmark records

Performance

Performance was measured on the reference configuration below. Results are intended to make the optimization reproducible but do not guarantee identical performance on every Arm-based system.

Reference configuration

Field	Value
Device / platform	vivo X300
CPU / accelerator	C1-Ultra, C1-Premium, C1-Pro (aarch64, 8 cores), CPU
OS	android - Android 16 / OriginOS 6
Runtime	LiteRT 0.9.0
Backend / delegate	XNNPACK + KleidiAI
Batch size	1
Precision	INT8 dynamic PTQ, per-channel symmetric weights, INT8 activations quantized dynamically per-tensor at runtime (stored FP32 on graph edges between kernels)
Runs	10 warmup + 50 measured

Performance results

Metric	Original / baseline	Arm-optimized	Improvement
p50 latency	174164.25 ms	8853.61 ms	19.67x
p90 latency	174412.33 ms	9117.33 ms	19.13x
RTFx (real-time factor)	0.07x	1.45x	19.67x
Model size	2914.22 MB	743.15 MB	3.92x smaller
Peak memory	6178.08 MB	3150.16 MB	1.96x less

Accuracy

Accuracy was evaluated using the same preprocessing, input resolution, and evaluation protocol described below. Where possible, the optimized model is compared against the original model under the same evaluation conditions.

Evaluation setup

Field	Value
Dataset	LibriSpeech ASR
Split	test-clean
Number of samples	2620
Metric(s)	WER (Word Error Rate), CER (Character Error Rate)
Evaluation runtime	LiteRT

Accuracy results

Metric	Original / baseline	Arm-optimized	Change
WER	2.88%	2.86%	-0.02 pp
CER	1.27%	1.26%	-0.01 pp

Accuracy was measured using the evaluation setup described above. Users should re-evaluate the model on their own data before production use.

Arm optimization approach

Arm optimized this model for efficient inference on Arm-based platforms using a hardware-aware conversion and validation flow.

For this release, Arm used:

Optimization area	Applied?	Notes
Model conversion	Yes	Re-authored the encoder/decoder as separate LiteRT multi-signature (encode/decode) modules and converted to .tflite
Quantization	Yes	AI Edge Quantizer dynamic_wi8_afp32 recipe - INT8 per-channel symmetric weights quantized offline plus INT8 activations quantized dynamically per-tensor at runtime; activations stored FP32 on graph edges; INT8 x INT8 -> INT32 kernel arithmetic; no calibration dataset required
Runtime/backend selection	Yes	XNNPACK + KleidiAI kernels selected for the LiteRT CPU delegate (4 threads)
Graph/runtime compatibility updates	Yes	Performed as part of the LiteRT export pipeline - the encoder/decoder were re-authored with explicit self-attention and cross-attention KV cache tensors and exported as a multi-signature (encode/decode) graph for autoregressive decoding
Accuracy validation	Yes	Compared against the original model or published baseline
Performance validation	Yes	Measured on the reference Arm platform

The goal of this process is to improve deployment characteristics such as latency, memory use, model size, and runtime compatibility while preserving the model's intended behavior. Detailed conversion scripts, calibration configuration, or backend-specific implementation details may be provided separately where appropriate.

Using this model

Install dependencies

Dependencies are declared in pyproject.toml, which ships with this repository. Resolve and install them into a local virtual environment with uv:

uv python install
uv sync --frozen

Run the example

uv run example.py

Note: The Python/uv example runs on AWS Graviton (Ubuntu arm64) to confirm runtime compatibility only, and is intended as a guideline for building an equivalent run script on Mobile CPU systems.

Expected input

Property	Value
Input shape	[1, 80, 3000] (log-mel spectrogram features)
Input type	float32
Input range	N/A (log-mel spectrogram, not a bounded range); raw audio input is 16000 Hz mono PCM in flac, wav, mp3, or ogg format
Preprocessing	Load audio (soundfile.read), downmix to mono, resample to 16000 Hz via linear interpolation if needed, compute an 80-bin log-mel spectrogram via the WhisperProcessor feature extractor (30s window, padded/truncated) to produce the [1, 80, 3000] float32 input tensor

Expected output

Property	Value
Output shape	N/A (variable-length autoregressive token sequence; per-step logits shape [1, 1, 51865])
Output type	text string (decoded from int64 token ids)
Postprocessing	Greedy argmax decoding per step using self-attention and cross-attention KV cache, then decode token ids to text (skipping special tokens) and strip surrounding whitespace

Intended use

This model is intended for developers evaluating automatic speech recognition workloads on Arm-based platforms. It is suitable as a reference implementation for benchmarking, prototyping, and integration exploration.

Limitations

• Performance depends on the target device, runtime version, backend/delegate support, memory configuration, and system load.
• Accuracy was evaluated on LibriSpeech ASR test-clean and may not generalize to all domains.
• This release preserves the original model's intended task and behavior, but users should validate it for their own application, data, and deployment environment.
• This repository is not a replacement for the original model documentation.

Additional notes

• Decoding is greedy only (no beam search), capped at 128 generated tokens per utterance, with language forced to English and task forced to transcription.
• Latency and memory figures are specific to the vivo X300 (aarch64, LiteRT + XNNPACK/KleidiAI, 4 threads); numbers on other Arm chipsets, thread counts, or Android versions may differ.
• WER/CER are computed after Whisper-style text normalization, which affects the numeric values (e.g., digit/number formatting); raw, unnormalized transcripts will show different error rates.
• Sample input: sample_input.flac is utterance 3575-170457-0005 of the LibriSpeech ASR corpus (test-clean split) by Panayotov et al., via OpenSLR (CC BY 4.0).

About this version

Original Model: openai/whisper-medium by OpenAI - Repository

Optimization/conversion: Arm-Optimized version for execution on Arm-based platforms.

Converted/optimized by: Arm

License: The Original Model and the Optimized Model are subject to Apache-2.0.

This repository contains a converted or optimized version of the Original Model (the “Optimized Model”). The Original Model has been converted or optimized as described above for execution on Arm-based platforms.

No retraining or fine-tuning of the Original Model was performed as part of the conversion or optimization. The conversion or optimization was not intended to change the Original Model’s behavior or intended use.

Original Model and Documentation

For information about the Original Model, including its development, training data, intended uses, limitations and other relevant information, please refer to the Original Model repository. Information in that repository was provided by the original developer or other third parties and, unless expressly stated otherwise, has not been independently verified by Arm.

Licenses and Third-Party Terms

Use of the Original Model and the Optimized Model is subject to the applicable licenses, usage restrictions and other terms identified above and in the relevant repositories. Publication of the Optimized Model does not grant any rights beyond those provided under the applicable license terms.

You are responsible for reviewing those terms and ensuring that your use of the Original Model and the Optimized Model is permitted.

Purpose of this Release

The Optimized Model is provided as a reference implementation to demonstrate and evaluate execution and performance on Arm-based systems. It is not a production-ready or supported solution.

Arm’s publication of the Optimized Model does not constitute an endorsement or certification of the Original Model or a representation that the Optimized Model is suitable for production use or any particular purpose.

To the fullest extent permitted by applicable law (i) the Optimized Model is provided “as is.” Arm makes no representations or warranties that the Original Model, the Optimized Model or their outputs are accurate, safe, secure, non-infringing, legally compliant, suitable for production use or fit for any particular purpose; and (ii) Arm will not be liable for any loss or damage arising from or in connection with the Optimized Model, its use or its outputs.

You are responsible for independently evaluating the Optimized Model, its outputs and its suitability for your intended use, including compliance with applicable legal, regulatory, safety and security requirements.

Arm does not commit to provide ongoing support, maintenance or updates for the Optimized Model. Any use of or reliance on the Optimized Model or its outputs is at your own risk.