- Home
- Skills
- Agents & Orchestration
- Alexa Voice Assistant in Rioplatense Spanish
Works with the AI Agents you already use
Alexa Voice Assistant in Rioplatense Spanish
Build Alexa skills backed by a language model that understand Argentine Spanish (voseo), say numbers, money and times as words, and answer inside Alexa's eight-second limit.
$15
Alexa Voice Assistant in Rioplatense Spanish
Example session with this skill installed
My Alexa-hosted skill is in Python and sends every sentence to a model, so even "poneme un temporizador" takes three seconds. Add a fast path so timers, alarms, reminders and the dollar rate are answered without the model, and make the spoken confirmations sound right in Argentine Spanish.
- Read your context and instructions
- Compiled the alexa voice assistant
Summary
Added a fast path in front of the model call. Timers, alarms, reminders and the dollar rate are now answered by pattern matching in well under a millisecond, and every figure is spoken as words. Everything else still goes to the model.
What I changed
Copied voz_es.py and via_rapida.py from the skill next to lambda_function.py, and added this function, called before the model in the chat handler:
from datetime import datetime
from via_rapida import via_rapida
from voz_es import decir_duracion, decir_hora, decir_pesos
def responder_rapido(frase, ahora, dolar_blue):
"""Spoken reply for an everyday command, or None to fall through to the model."""
a = via_rapida(frase, ahora) # ahora = the user's local time, not the Lambda clock
if not a:
return None
if a["tipo"] == "temporizador":
return f"Listo, {decir_duracion(a['segundos'])}." # then create the Alexa timer
if a["tipo"] == "recordatorio" and "fecha_hora" in a:
c = datetime.fromisoformat(a["fecha_hora"])
dia = "hoy" if c.date() == ahora.date() else "mañana"
return f"Dale, {dia} a {decir_hora(c.hour, c.minute)}." # then create the reminder
if a["tipo"] == "recordatorio":
return f"Dale, en {decir_duracion(a['en_minutos'] * 60)} te aviso."
if a["tipo"] == "dolar":
return f"El blue está a {decir_pesos(dolar_blue)}." # figure read from dolarapi.com
return None
Result
Run with the local time set to a Friday at 15:00:
| What the user says | Path | Spoken reply |
|---|---|---|
| poneme un temporizador de diez minutos para los fideos | fast path | Listo, diez minutos. |
| despertame a las siete y media | fast path | Dale, mañana a las siete y media de la mañana. |
| recordame en veinte minutos sacar la pizza | fast path | Dale, en veinte minutos te aviso. |
| recordame a las 9 de la noche que saque la basura | fast path | Dale, hoy a las nueve de la noche. |
| a cuánto está el blue | fast path | El blue está a mil quinientos setenta y cinco pesos. |
| contame un chiste | model | (falls through to the model) |
| no, lo que quería era otra cosa | model | (falls through to the model) |
Things the skill's helpers got right that a plain pattern would miss:
- "despertame a las siete y media", said at three in the afternoon, is read as tomorrow morning because it is a wake-up alarm. A reminder "a las 7" at the same hour is read as 19:00 today.
- The reminder text keeps its accents, so the voice reads it back correctly.
- "1575" is spoken as "mil quinientos setenta y cinco pesos". A figure written as "1.575" is never sent to the voice.
- A sentence starting with "no" is a correction, so it goes to the model with the conversation so far.
Tests
$ python -m unittest test_voz
..................................
Ran 34 tests in 0.582s
OK
Next steps
- Wire the timer and reminder actions to the Alexa Timers and Reminders APIs (the action dicts already carry
segundos,fecha_horaoren_minutos, and the label). - Add a route for each command your users repeat every day; the guide
references/rioplatense.mdlists the voseo forms to match. - Run the phrases in
assets/frases-de-prueba.txtin the Alexa developer console and check the response times.
Connects securely to your tools. The creator never sees your data.
About this skill
A voice assistant that calls a language model has problems a chat app does not. Alexa cuts the request off after about eight seconds. The input is transcribed speech with errors. The output is heard, so symbols and digit groups come out wrong. In Argentina and Uruguay there is one more: people speak with voseo ("poné", "fijate", "recordame") and local vocabulary that neutral-Spanish patterns miss.
This skill gives your coding agent the working solution to all of them.
Where it comes from
It comes out of Cerebro, my open source Alexa assistant in daily use (MIT licence, free on GitHub at github.com/alfregarrone/cerebro). Cerebro is a finished assistant you install as it is. This skill is a separate package made for coding agents: the voice layer rewritten as small tested modules, plus the workflow, templates and guides an agent needs to build the same behaviour into your own skill. None of the files in this package are in the repository.
What your agent gets
- A step-by-step workflow for building or improving the skill: interaction model, handler order, time budget, fast path, model calls, spoken output, testing.
- voz_es.py: normalises transcribed speech, reads numbers ("dos mil quinientos", "1.480,50"), durations ("hora y media") and clock times ("a las ocho menos cuarto"), understands spoken confirmations ("sí, dale", "no gracias"), and says numbers, money and times back as words with correct agreement ("veintiún minutos", "las siete y media de la tarde").
- via_rapida.py: answers everyday commands (timers, alarms, reminders, dollar rate, weather, lights) with patterns in well under a millisecond, with no model call.
- carrera.py: races several model calls inside a deadline and rests models that are out of quota. Works with any provider.
- acciones.py: the "model proposes, code executes" protocol. Splits the reply into speech and proposed actions, validates them against a closed list, and holds risky ones for spoken confirmation. The model cannot mark its own action as confirmed.
- Templates: a catch-all interaction model for es-US with Rioplatense sample values, a system prompt written in Rioplatense Spanish, and a phrase battery for end-to-end checks.
- Reference guides: voseo verb forms and vocabulary, recogniser slips, the request flow, and key-free public APIs for Argentine live data (dollar rates, holidays, inflation, weather, news).
Good to know
- Standard-library Python, 34 unit tests included.
- The scripts hold no API keys and make no network calls.
- The helpers also fit voice assistants outside Alexa, such as a Telegram bot with voice notes.
Not covered
Publishing and certification, account linking and audio streaming. The code is Python; the reference guides apply to any language.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires Python for provided scripts. Does not include account linking, audio streaming, or Amazon certification workflows. Logic is specifically tuned for Rioplatense Spanish.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 days ago
- Software
Python 3.8+
Permissions
The bundled scripts are standard-library Python and make no network calls. The agent runs the unit tests in a terminal and copies the modules into your project. Network access applies only when the agent adds live-data actions to your Alexa skill, using the public key-free APIs documented in references/argentina-data.md (dolarapi.com, api.argentinadatos.com, open-meteo.com, news.google.com).