Editing Openai/691bd6f8-a5e8-8005-9752-1925a60e6a18 (section)

=== 1) How Alexa’s “personality” was created historically ===
# Brand definition + scripted voice style. Amazon defined a consistent brand/personality (approachable, efficient, trustworthy, sometimes witty) and produced voice-scripts, joke banks, canned micro-responses, and style guidelines so the assistant would behave consistently across many situations. That style guide governs things like humor, how Alexa answers “I love you,” small talk, etc. Developer Portal Master<ref>{{cite web|title=Developer Portal Master|url=https://developer.amazon.com/en-US/alexa/branding/alexa-guidelines/communication-guidelines/brand-voice|publisher=Developer Portal Master|access-date=2025-11-20}}</ref>
# Pipeline and TTS + voice recordings. Early Alexa combined classical speech tech: wake-word detection and ASR trained on huge speech corpora (many recordings gathered/annotated via vendors), an NLU/dialog manager that mapped intents to actions or canned responses, and a TTS engine (Amazon Polly / neural TTS) that rendered the chosen response in a consistent voice. Real human recordings and later neural TTS improvements shaped the timbre and prosody. Encyclopedia Britannica<ref>{{cite web|title=Encyclopedia Britannica|url=https://www.britannica.com/technology/Amazon-Alexa|publisher=Encyclopedia Britannica|access-date=2025-11-20}}</ref>
# Supervised learning + human labels. Amazon trained ASR, NLU intent classifiers, slot taggers, and dialog-state components using large labeled datasets and manual annotation. Supervised and semi-supervised learning were central because manual labels gave reliable behavior at scale. About Amazon<ref>{{cite web|title=About Amazon|url=https://www.aboutamazon.com/news/devices/how-our-scientists-are-making-alexa-smarter|publisher=About Amazon|access-date=2025-11-20}}</ref>
# Research + iterative human feedback. Programs like the Alexa Prize and Amazon Science research (multi-turn tracking, context, retrieval, safety) supplied new algorithms, evaluation data and user ratings that helped Alexa produce more coherent, engaging multi-turn speech. Amazon also used in-house writers and curated content for specialized features (e.g., interactive stories). Amazon Science<ref>{{cite web|title=Amazon Science|url=https://www.amazon.science/alexa-prize|publisher=Amazon Science|access-date=2025-11-20}}</ref>
# Human-in-the-loop & continuous improvement. Every day Alexa interactions (anonymized and subject to privacy settings) produced training signals; human reviewers labeled or rated outputs and were used to tune models, refine safety filters and adjust the assistant’s conversational habits. About Amazon<ref>{{cite web|title=About Amazon|url=https://www.aboutamazon.com/news/devices/how-our-scientists-are-making-alexa-smarter|publisher=About Amazon|access-date=2025-11-20}}</ref>