Why this matters
Client calls, interviews, team meetings and voice notes hold decisions that nobody writes down, or that get written up a day late from memory. Voice AI captures all of it as text, then turns that text into minutes, action items and themes.
Every tool in this course, from Wispr Flow dictation to Speak AI, meeting notetakers and ChatGPT meeting summaries, starts with the same step: speech-to-text. Its mistakes are not random. They land on names, numbers and jargon, the words that matter most, so knowing how it works tells you exactly what to check.
What you’ll build
You follow one client call recorded by Fairbank HR Consulting (made up) from sound to transcript, spot a transcription error, watch speaker diarization split the call by voice and name the speakers, then turn the transcript into a summary, action items and themes.
What you need
- Nothing but a browserIt is a short explainer of diagrams you click to open. No account, no recording and nothing to set up.
Concepts
3 ideas you’ll use in the practice.
Speech-to-text and word error rate
A speech recognition model writes the most likely words with timestamps, then adds punctuation. Accuracy is measured as word error rate: 5 wrong words in 100 is 5%. Clean audio, one voice at a time and a close microphone matter more than which tool you use.
Speaker diarization: who said what
Diarization groups stretches of speech that sound alike and labels them Speaker 1, Speaker 2. It does not know names, so you add them. Crosstalk and similar voices on one microphone are where it merges or swaps people.
From transcript to work
A summary is the short version. Action items need an owner (from the speaker labels) and a date (from the words). Thematic analysis groups what many people said into themes with real quotes. Check names, dates and quotes against the audio before anything goes to a client.
Practice
Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.
Do it for real
Now repeat it in your own account. Tick each step as you go.
Quick check
3 questions to check you’ve got it.
1. Where do speech-to-text mistakes usually land?
- Short words like "and" and "the"
- Names, numbers, dates and jargon
- Only on the first minute
Show the answer
Names, numbers, dates and jargon. Rare words are hardest to recognize, and they are the ones that matter most. Check them first.
2. What does speaker diarization do?
- Translates the call
- Works out who spoke when and splits the transcript by voice
- Removes background noise
Show the answer
Works out who spoke when and splits the transcript by voice. It groups speech by how it sounds. You still put the names on the groups.
3. An AI summary says a client "hates the new pricing". What should you do before it goes in a report?
- Trust it, the AI read the whole call
- Find the quote and listen to it
- Delete the theme
Show the answer
Find the quote and listen to it. A theme is a claim about your customers. Listening to its quotes is how you catch a transcription error before it becomes a wrong conclusion.
Wispr Flow: Answer Emails and Write AI Prompts by Voice
You know how talk becomes text. Next, put it to work on your own writing: answer emails and write AI prompts by voice with Wispr Flow.
Start module 3 →