iPhone Voice Transcriptions Goes Horribly Wrong

A Scottish grandmother received an obscene and insulting message thanks to a complete fail by Apple's voice transcription. It completely misunderstood a voicemail left by a local car dealership.

The call was made to Louise Littlejohn by a worker at the garage who intended to invite her to a promotional event for a new car. The call went to voicemail so he left a message. Apple then "transcribed" the call with an AI-powered system and delivered a text message, something that's meant to save users time rather than have to listen to the message.

According to the BBC, which listened to the audio of the call, the message actually said:

"Hi Mrs Littlejohn, it is [redacted] here from Lookers Land Rover in Lanarkshire. I hope you are well. Just a wee call to see if you have received your invite to our new car [inaudible] event that we do have on between the sixth and tenth of March. Just a wee call to see if it is something you were looking to come along to, and to see if we can confirm an appointment slot that would be suitable for yourself. If it is something you would be interested in, feel free to give me a call on [redacted], ask for myself [redacted] [inaudible]. Thank you."

AI Translates with Insulting Message

Unfortunately the message Littlejohn received was very different, not only bearing little relationship to the actual wording, but including some unfortunate wording:

"Lots of [redacted] CT from work line over line trail of you will just be told to see if you have received an invite on your car if you've been able to have s@x and not what should we call and just keep trouble with yourself that'd be interesting you piece of sh!t give me a call [redacted] customer [redacted] will have to help us thank you."

Accent Not The Issue

A speech technology expert interviewed by the BBC said the caller's Scottish accent was unlikely to be the cause of the significant failure.

Instead, he believed the main problems were the poor quality of the call and the background noise. Perhaps surprisingly, Professor Peter Bell also pointed to the caller reading from a script, meaning his speech pattern was more formal and less natural than normal. (Source: bbc.co.uk)

Neither Apple nor the car dealership has commented on the incident. (Source: thenational.scot)

How AI Translation Works

Apple's voicemail transcription feature uses AI-based speech recognition to convert voicemail audio into written text. When someone leaves a voicemail, the audio is processed by machine learning algorithms that attempt to identify words and structure sentences. This process may occur on the device or in the cloud, depending on device settings and system capabilities.

Limitations of AI Transcription

AI transcription is not perfect and struggles with several factors. Poor audio quality, background noise, overlapping speech, and unclear enunciation can lead to severe errors. Scripted or overly formal speech, like in this case, may also confuse the system because it deviates from everyday conversational patterns that AI models are trained on.

Comparison with Other Services

Apple is not the only company offering voicemail transcription. Google Voice and Microsoft Teams also use speech-to-text technology.

In some independent tests, Google tends to offer more accurate transcriptions, especially with diverse accents and background noise. However, all systems are prone to failure in edge cases, and none are completely reliable.

Accent and Dialect Handling in AI

While many assume that regional accents are the biggest problem for AI transcription, experts say otherwise. Most modern systems are trained on a wide variety of accents. The real challenge is when those accents are combined with poor audio or unnatural speaking patterns, such as reading a script. That combination often leads to bizarre or offensive output.

User Privacy Concerns

Errors like this raise serious questions about user privacy and control. If AI systems can generate offensive or misleading messages, should users be warned before delivery? Is the content stored or analyzed further? These questions highlight the importance of transparency in how voice data is processed, especially by companies like Apple.

The Growing Role of AI in Everyday Communication

AI tools are becoming a central part of daily communication, from voicemail transcription to predictive text and real-time translation. While these technologies offer convenience, they also carry risks. A single misinterpretation, like in this case, can cause embarrassment, confusion, or worse. As reliance grows, so does the need for accuracy and accountability.

AI Fails in Other Domains

This incident is not unique. AI has produced offensive or incorrect content in many other areas, including chatbots making racist remarks, facial recognition software misidentifying people, and predictive text systems suggesting inappropriate language. Each failure underscores the importance of human oversight and more rigorous testing before deployment.

What's Your Opinion?

Have you had similar problems with auto-transcription? Is this just an unfortunate but rare incident? Should Apple have fixed such problems by 2025?