Punctuation and layout
The rules turn spoken commands into commas, periods, new lines, and new paragraphs, and can apply capitalization changes.
CareCompile speech program
Dictation for regional Spanish
Dictation combines two distinct problems: recognizing how a person speaks and applying commands such as “coma,” “punto,” “nueva línea,” or “no perdón.” Vocero has a text-based rule layer; its own speech recognition remains pending.
In development for internal use. Our own model is not trained yet.
01 / VOCERO
The component already implemented
The browser demo runs deterministic rules on what a person types. It does not activate the microphone or send the text.
Correction and undo
Correction, deletion, repetition, and undo commands modify the text through traceable steps.
Ambiguities
The engine distinguishes uses such as “coma” in a clinical context or “punto” in a reference and flags cases that require review.
02 / VOCERO
What the regional plan must cover
The proposed adaptation starts from language features and situations that a general system may handle unevenly.
Voseo and vocabulary
The plan includes forms such as “vos tenés” and “decime,” words such as “pisto,” “cipote,” and “chunche,” and regional proper names.
Speech variation
The planned set should represent departments, ages, and genders, as well as features such as syllable-final /s/ aspiration.
Meeting conditions
Evaluation should cover several speakers, switching between Spanish and English, and phone-quality audio.
03 / VOCERO
Speech recognition is future work
The rule layer receives text. Before it, a speech model would need to recognize the words and tag the commands.
Candidate model
The program will evaluate open speech models and tune one on regional audio only after establishing a baseline.
Consented data
Dictation scripts would include commands, numbers, dates, and regional names read by people who expressly authorize the use.
No attributed result
Today's working dictation uses another engine. The rules demo must not be interpreted as output from Vocero's own model.
04 / VOCERO
How dictation would be evaluated
Vocero-Bench is designed, but it does not yet contain recordings or results.
Recognized text
The plan measures substitutions, deletions, and insertions through word and character error rates, including regional proper names.
Punctuation and commands
It would also measure punctuation, command order, false activations, and commands written as words instead of executed.
Results by country
The planned comparison separates each country and uses different speakers for training and testing. No system has been measured yet.
Explore the program
Related pages
From a meeting to reviewable minutes
Explore Vocero's design for transcribing meetings and producing minutes with decisions, tasks, and source citations. In development; the demo is synthetic.
Explore the scopeA counterpart for Salvadoran voices and texts
CareCompile seeks an institutional counterpart in El Salvador to explore consented speech and text data, limited purpose, and independent evaluation.
Explore the scopeArchitecture and evaluation before claims
Review Vocero's proposed architecture, training plan, and Vocero-Bench design for measuring speech, dictation, and minutes by country. There are no results yet.
Explore the scopeEL SALVADOR / CARECOMPILE
Designate an institutional counterpart in El Salvador
We are looking for an institution willing to discuss regional meetings, voices, and texts. We explain purpose, consent, and governance before recording or receiving data.