Research

Compare VoicePing MT v0.1 and seven alternatives on 100 English-to-Japanese rows, with model-judged quality, observed latency and evaluation limits.

Compare VoicePing Diarization v0.1 with NeMo, pyannoteAI, AssemblyAI and Deepgram on a 42-file multilingual benchmark, with data and metric limitations.

Compare five speaker identification models on held-out speakers, short clips and unknown-speaker rejection, with a separate 900-speaker diagnostic.

Compare eight open OCR models on 900 pages in five languages, with accuracy, completion, GPU latency, throughput and memory results.

Compare 13 open models, GPT-5.6 Luna and GPT-5.4 for meeting summaries. See model-by-model quality, response speed, valid structure and GPU results.

Four OCR APIs, 20 real images and 240 requests: compare accuracy, speed and failures for Google, Azure, Mistral and OCR.Space, including failed outputs.