Note Master dataset: every take from the capture app. Subsets below are copies of its rows.

Note Subset `code_switch`: English sentences with everyday Hebrew words and Israeli names written in Latin letters, read aloud by one English speaker living in Israel.

Note Subset `tech`: Dictation-style English sentences dense in software, DevOps, cloud, self-hosting and AI vocabulary (product names, CLI tools, acronyms, model names), read aloud by one speaker.

Note Subset `ai_interactions`: Instructions dictated to AI coding agents (Claude Code, Codex and others) and the git and GitHub workflow around them, read aloud by one speaker: subagents, worktrees, compaction, plan mode, hooks, MCP servers, pull requests, rebases.

Note Subset `linux`: Dictation-style English sentences about Linux desktop and server administration (KDE Plasma, GNOME, systemd, storage, networking, homelab), read aloud by one speaker.

Note Subset `smart_home`: Dictation-style English sentences about home automation (Home Assistant, Zigbee, Z-Wave, Matter, Thread, ESPHome, MQTT and device brands), read aloud by one speaker.

Note Subset `personal`: English sentences built around names a speech model gets wrong for one speaker: his own name, his company and his project names (Daniel Rosehill, DSR Holdings, Carrot Cake AI, the podcast My Weird Prompts and its hosts Corn and Herman Poppleberry), read aloud by that speaker.

Note Subset `homophone`: English sentences whose context settles which spelling of a homophone is meant (their/there, principal/principle, and 58 more sets), read aloud by one speaker.

Note Subset `yiddish`: English sentences with Yiddish words written in Latin letters (shul, daven, cholent, mensch, freier, tachles), the way English speakers in Israel use them, read aloud by one speaker.

Note Subset `religious`: English sentences with Jewish religious vocabulary written in Latin letters: Shabbat and the festivals, prayer and the synagogue, Torah portions, Talmud and halacha, kashrut, and life-cycle events, read aloud by one speaker.

Note Subset `general`: Plain, everyday English sentences (emails, messages, general prose and homophone pairs) read aloud by one speaker, with no Hebrew or Yiddish words, Israeli names or tech vocabulary.

Note Subset `noisy`: Every take from the main dataset recorded in noise, whatever its category: outdoors, in a noisy room, with traffic, other people talking, background noise or a crying baby.

Note Task subset `disfluency`: Experimental.

Note Task subset `self-corrections`: One speaker correcting himself mid-sentence ("send it Tuesday, no, Wednesday").

Note Task subset `asides`: One speaker dictating a sentence and, partway through, saying something to someone in the room ("come in, Hannah").

Note Task subset `hard-asr`: English sentences speech recognition finds hard, read by one speaker: Harvard sentences (IEEE 1969, public domain), tongue twisters, homographs, garden-path sentences, rare words, dense numbers and units, unusual proper nouns.

Note Task subset `brand-names`: Natural sentences with company, product, AI model and open-source project names (Anthropic, OpenRouter, Amazon Bedrock, Qwen, Kubernetes, Nginx) read by one speaker, written in each name's official form.

Note Task subset `acronyms`: Natural sentences with acronyms read by one speaker, both spelled out (API, SSH, HTTPS) and said as words (NASA, JSON, YAML), each in its standard written form.

Note Task subset `irish-english`: Everyday sentences in Irish English read by one speaker born in Ireland: vocabulary (craic, grand, giving out, press, messages), grammar (I'm after doing it, so it is, ye), Irish-language words used in English (Taoiseach, Dáil, Gardaí, sláinte) and Irish names and places (Siobhán, Tadhg, Dún Laoghaire), in standard Irish spelling with fadas.

Note Task subset `repetitions`: Experimental.

Note Task subset `singing`: Experimental.

Note Task subset `weird-voices`: Experimental.

Note Task subset `currency`: One speaker saying amounts of money the natural way ("50 shekels", "a dollar fifty", "6 pounds 50").

Note Task subset `expanded-acronyms`: Chat and business shorthand said as letters by one speaker ("by E O D", "F Y I", "I M O") with the transcript writing it out in full ("by end of day", "For your information", "In my opinion").

Note Import: the earlier code-switching recordings, in the master dataset as `ehms_*` rows.

Note Blend dataset for run `r7-all` (kitchen sink: English-Hebrew code-switching plus tech vocabulary, accent and general accuracy): the exact training mix of Hebrish/Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching-All-Subsets.

Note Blend dataset for run `r7-general` (no Hebrew: tech vocabulary, accent and general accuracy): the exact training mix of Hebrish/Parakeet-TDT-0.6B-v2-Tech-And-Accent.