Benchmarking Commercial ASR Systems on Code-Switching Speech: Arabic, Persian, and German
Summary
This paper presents a benchmark evaluating five commercial ASR systems on code-switching speech across Arabic-English, Persian-English, and German-English pairs, using a two-stage pipeline to select 300 samples per pair and assessing performance with WER and BERTScore. ElevenLabs Scribe v2 achieves the lowest overall WER (13.2%) and highest BERTScore (0.936), with public dataset available.
View Cached Full Text
Cached at: 05/20/26, 08:23 AM
# Benchmarking Commercial ASR Systems on Code-Switching Speech: Arabic, Persian, and German Source: [https://arxiv.org/abs/2605.19069](https://arxiv.org/abs/2605.19069) [View PDF](https://arxiv.org/pdf/2605.19069) > Abstract:Code\-switching \-\- the natural alternation between two languages within a single utterance \-\- represents one of the most challenging and under\-studied conditions for automatic speech recognition \(ASR\)\. Existing commercial ASR benchmarks predominantly evaluate clean, monolingual audio and report a single Word Error Rate \(WER\) figure that tells practitioners little about real\-world multilingual performance\. We present a benchmark evaluating five commercial ASR providers across four language pairs: Egyptian Arabic\-\-English, Saudi Arabic \(Najdi/Hijazi\)\-\-English, Persian \(Farsi\)\-\-English, and German\-\-English\. Each dataset comprises 300 samples selected by a two\-stage pipeline: a heuristic filter scoring transcripts on five structural code\-switching signals, followed by a GPT\-4o and Gemini 1\.5 Pro ensemble scoring candidates across six linguistic dimensions\. This pipeline reduces LLM scoring costs by approximately 91\\% relative to exhaustive scoring\. We evaluate the systems on both WER and BERTScore, arguing that BERTScore is a more reliable metric for Arabic and Persian pairs where transliteration variance causes WER to penalise semantically correct transcriptions\. ElevenLabs Scribe v2 achieves the lowest WER across all four language pairs \(13\.2% overall; 13\.1% on Egyptian Arabic\) and leads on BERTScore \(0\.936 overall\)\. We further demonstrate that difficulty\-stratified analysis reveals performance gaps masked by aggregate averages, and that BERT embedding projections confirm semantic proximity between reference and hypothesis despite surface\-level script differences\. The benchmarking dataset is publicly available at[this https URL](https://huggingface.co/datasets/Perle-ai/ASR_Code_Switch)\. ## Submission history From: Sajjad Abdoli \[[view email](https://arxiv.org/show-email/5ada2abb/2605.19069)\] **\[v1\]**Mon, 18 May 2026 19:50:44 UTC \(748 KB\)
Similar Articles
Can Voice Agents Handle Bilingual Customers? Benchmarking Frontier ASR on Code-Switched Speech
ServiceNow AI releases a benchmark and dataset for evaluating automatic speech recognition (ASR) on code-switched speech across four language pairs (Spanish-English, French-English, Canadian French-English, German-English) in enterprise HR and IT scenarios, finding that current frontier ASR models still struggle with code-switching, leading to higher error rates.
Benchmarking Human and Automatic Speech Recognition of Diverse Speech: Initial Results
This paper compares state-of-the-art ASR systems to human listeners on recognizing diverse Dutch speech, finding that ASR systems match or exceed human performance in some cases, with Google Telephony leading. It highlights the impact of speaker age, regional accents, and test set selection on benchmarking conclusions.
Code-Switching Information Retrieval: Benchmarks, Analysis, and the Limits of Current Retrievers
Researchers introduce CSR-L and CS-MTEB benchmarks showing that code-switching queries degrade IR system performance by up to 27%, revealing embedding-space divergence that current multilingual techniques cannot fix.
Transcribing Children's Speech: ASR Performance and Obtaining Reliable Orthographic Transcriptions
This paper evaluates nine ASR models (Whisper, Parakeet, Wav2Vec2) on Dutch child speech datasets JASMIN and DART, finding that fine-tuned Whisper-medium achieves the best performance (WER 5.54% on JASMIN, 70.37% on DART). It also proposes a selection method to automatically identify correctly pronounced utterances with high precision, reducing the need for manual verification.
What Counts as an Error? Dual-Reference Benchmarking for Atypical ASR
This paper proposes a dual-reference benchmarking approach for atypical ASR, using both verbatim and intended transcriptions to evaluate 11 ASR models on stuttered speech, highlighting the importance of selecting the appropriate reference depending on the use case.