They then annotate the recordings they transcribe and feed it back into Alexa's "brain" to improve its understanding and fill in any remaining blank spaces with regards to human speech.
According to research from a year ago, one in 10 people has one or more smart speakers in their home, with Amazon's range of Echo smart devices the market leader. In one shift, the analysts will listen in on as many as 1,000 clips, which the firm says is minuscule considering the tens of millions of people who own the systems.
Because of this, Amazon workers often catch weird things: actual examples range from a woman singing badly off-tune in the shower to children crying for help to what was believed to be a domestic sexual assault, the two sources told Bloomberg.
But Amazon said it takes "security and privacy of our customers' personal information seriously".
A spokesperson for the trillion-dollar company said that "an extremely small sample of Alexa voice recordings" are analysed by staff. The marketing materials for the Echo claim "Alexa lives in the cloud and is always getting smarter", only hinting in the lengthy Alexa FAQ that "We use your requests to Alexa to train our speech recognition and natural language understanding systems". Even so, it's concerning to think that private conversations could be listened to by employees if Alexa is accidentally woken after misinterpreting regular speech as its wake word, "Alexa".
As a result, workers occasionally hear things they shouldn't. Having raised the issue with the company, they say they were told that it wasn't Amazon's place to interfere. The workers also have no way of identifying who they are listening to as any personal or account information is removed. Employees can discuss what they hear with other employees in an internal chat service, and they can share clips they have trouble interpreting-though the report also mentions files being shared simply because they are "amusing".
Amazon is still largely trusted by consumers, but that could change with too many privacy snafus.
'All information is treated with high confidentiality and we use multi-factor authentication to restrict access, service encryption and audits of our control environment to protect it'.
"By default, Echo devices are created to detect only your chosen wake word (Alexa, Amazon, Computer or Echo)".
Ironically, at Google, employees are only allowed access to a very small and distorted audio clip that is completely unidentifiable. Apple uses a similar system of human workers to improve Siri, but it doesn't include any identifying data with the recordings. In the Alexa Privacy Settings there is a "Manage How Your Data Improves Alexa" section where you can decide whether your voice recordings are used to "help develop new features" and/or "help improve transcription accuracy".