Generic dictation mishears professional vocabulary — "metoprolol" becomes word salad, "voir dire" becomes "vwa deer". Pithflow's specialty term packs teach the transcription your field's vocabulary, so professional dictation comes out clean the first time.
Speech-to-text models are trained mostly on everyday speech, so they default to common words. When a doctor says "metoprolol", a generic model hears something rare and guesses something common — "met a pro lol". Same failure for legal Latin ("res judicata"), software tools ("Kubernetes", "PostgreSQL"), and finance vocabulary ("EBITDA"). The fix isn't a better microphone; it's telling the transcription what vocabulary to expect.
Enable a pack in the Dictionary tab (or pick your field at signup) and Pithflow biases every transcription toward that vocabulary BEFORE the audio is decoded — the model expects the terms, so it recognizes them. A second pass repairs near-misses: if a mangled rendering slips through, the AI cleanup compares it against the pack and restores the correct form. Your personal dictionary terms always take priority on top.
• Medical (English) — top-prescribed drugs, clinical exam vocabulary, diagnoses, anatomy, procedures. • Médico (Español) — vocabulario clínico y fármacos for Spanish-speaking clinicians, including Mexican and LATAM usage. • Software & engineering — tools, frameworks, infrastructure, CS concepts that normally get mangled. • Legal (English) — Latin phrases, litigation and contract vocabulary. • Legal (Español) — amparo, derecho procesal, civil y notarial mexicano. • Finance & accounting — reporting terms, instruments, filings, acronyms. • Science & research — lab techniques, genetics, statistics, publishing vocabulary. • Mental health & therapy — modalities, diagnoses, clinical documentation. • Education & teaching — IEP/504 documentation, pedagogy, literacy, assessment. • Construction & trades — framing, concrete, MEP, permits and project paperwork. • Marketing & sales — SEO/PPC metrics, funnels, attribution, platform names.
Spanish-speaking professionals can enable both the English and Spanish medical packs and code-switch freely — "paciente con hipertensión, started him on metoprolol" transcribes each part correctly. Auto-detect handles the language per dictation; the packs handle the vocabulary in both.
Packs cover the high-yield vocabulary of each field, not every regional brand name or niche term. For the rest: the personal Dictionary (add a term once, it's recognized from then on) and Fix-a-word in History — spot a miss, type the correction, one click. Both feed the same recognition biasing as the packs.
Term packs improve transcription quality; they don't change data handling. Pithflow's current tiers are not HIPAA-eligible — an Enterprise tier with BAA is planned for Q4 2026. For workflows that require HIPAA compliance today, wait for Enterprise before dictating PHI.