Birthday Speech Generator Methodology

How We Assemble a Speech

Every speech is built from four optional components: an opening line, one or two body paragraphs, an optional anecdote bridge, and a closing toast. We pick one template from each bucket at random — filtered to the tone and relationship you chose — then substitute {name}, {age}, and optional {anecdote} into the text. The generator returns three unique variations so you can pick the phrasing that sounds most like you.

short speech  = opening + (optional anecdote bridge) + closing
medium speech = opening + body + (optional anecdote bridge) + closing
long speech   = opening + body + (optional anecdote bridge) + bodyTwo + closing

opening, body, bodyTwo chosen from OPENINGS[tone][relationship] etc.
bodyTwo is deduplicated against body so the two paragraphs differ.
wordCount    = token count on whitespace
speakingTime = wordCount / 130 wpm

Tone × Relationship Matrix

The template library is indexed by four tones and six relationships — 24 combinations, each with 2–3 opening variations, 2–3 body paragraphs, 2 anecdote bridges, and 3–4 closing lines. That keeps every generated speech feeling hand-written instead of Mad-Libs.

Length Targets

We set hard word-count windows for each length so the generator never produces something that runs long. Speaking time is derived from a 130 words-per-minute rate — the conversational-English average used by Toastmasters and major public-speaking coaches.

Sources

• Toastmasters International guidance on toast length and pacing for social occasions (the 1–2 minute standard toast).
• Conversational speaking-rate research: 130 words per minute is the widely cited average for English-language public speaking, used as the baseline for audiobook narration and presentation training.
• Party Genius AI voice style guide (docs/business/VOICE_STYLE_GUIDE.md) for tone calibration — heartfelt, funny, formal, and casual templates were written against explicit voice rules.
• Relationship category frame derived from Hallmark and American Greetings category taxonomies for birthday messaging (parent, friend, partner, sibling, coworker, general).

Edge Cases & Limitations

The generator is template-driven, not AI-generative — that is intentional. It gives you consistent quality, no hallucinations, and zero risk of awkward phrasing. The tradeoff is that with anecdotes, the bridge sentence is fixed even though your story is unique; we recommend reading the bridge, then rewriting it in your own words if the hand-off feels stiff.

Long speeches pick two distinct body paragraphs, but if a particular tone × relationship slot only has one body template, the second pick will reuse the same paragraph. This is rare and documented in the template catalog.

We do not currently generate culturally specific openings (e.g., Hindu blessings, Quinceañera toasts, Bar/Bat Mitzvah structures). Roadmap item — if you need one, write your opening line and paste it in front of the generated body.