According to a recent LinkedIn post from Perle, the company has developed what it describes as a benchmark for code‑switched speech recognition that focuses on real-world multilingual usage. The post indicates that the evaluation spans 1,200 hand-verified utterances mixing Egyptian Arabic, Saudi Arabic, Persian, and German with English, and compares seven commercial automatic speech recognition systems using WER and BERTScore.
The post suggests that performance for German-English combinations appears relatively mature, while Arabic-English and Persian-English scenarios expose significant weaknesses in most evaluated systems. For investors, this emphasis on under-served multilingual markets may position Perle as a specialist in evaluation tooling for ASR and LLMs, potentially creating opportunities in regions where code-switching is common and existing commercial offerings are less reliable.
By publishing benchmark results, a paper, and an open dataset, the post implies that Perle is pursuing an ecosystem approach that could increase its visibility among ASR developers and enterprise buyers. The invitation to add more models to the leaderboard may help Perle deepen relationships with vendors and customers in multilingual markets, which could translate into future monetization through evaluation services, partnerships, or related software tools.