What Happens to Your Voice Data (Hint: We Don't Keep It)

Your voice is uniquely yours. It carries biometric characteristics that can identify you with high accuracy. It reveals your emotional state, your health, your accent, and countless other personal details. When you use a voice assistant, you are sharing something far more intimate than text input. At Air, we take this responsibility seriously.

Voice Data Is Biometric Data

Before we explain what happens to your voice data, it is important to understand why voice privacy matters more than you might think. Your voice is biometric data in the same category as your fingerprint or facial geometry. It can be used to identify you, track you, and profile you in ways that text never could.

Voice contains information beyond just the words you speak. Acoustic analysis can reveal your gender, approximate age, emotional state, health conditions, native language, and geographic origin. Your unique vocal characteristics, including pitch patterns, speaking rhythm, and pronunciation habits, create a voiceprint that can identify you as reliably as a fingerprint.

Companies that collect and retain voice data have access to all of this information, whether they choose to use it or not. The mere existence of a voice archive creates risk, both from potential misuse by the company and from data breaches that could expose your biometric information to malicious actors.

The Complete Journey of Your Voice Data

When you speak a command to Air, your voice takes a brief journey through our infrastructure before returning to you as transcribed text. Here is exactly what happens at each step of the process.

First, your Mac's microphone captures your voice and converts the analog sound waves into digital audio data. This raw audio exists only in your Mac's memory at this stage. We apply local audio processing including noise reduction and normalization to improve transcription accuracy. This processing happens entirely on your device.

Second, the processed audio is encrypted and transmitted to our transcription servers. We use industry standard TLS encryption for all data in transit. Your audio travels over secure connections that prevent interception or eavesdropping.

Third, our servers receive the encrypted audio and forward it to our speech recognition system. The transcription process analyzes the audio waveform and converts it to text. This is the only point where your audio exists on any machine other than your own Mac.

Fourth, the transcribed text is returned to your Mac and displayed in the Air interface. At this point, the audio processing is complete.

Fifth, and this is the crucial part, we delete the audio data immediately after transcription is complete. There is no archive, no backup, no retention period. The audio ceases to exist on our servers within seconds of the transcription completing.

The entire round trip takes approximately 2 seconds, and your voice data exists on our infrastructure for only a fraction of that time.

What We Do Not Do With Your Voice Data

Many voice assistant companies retain recordings indefinitely for various purposes. They use your voice to train their speech recognition models, improve their services, conduct research, and sometimes even have human employees review recordings for quality assurance. We do none of these things.

We do not train our speech recognition models on your voice data. Our models are trained on licensed datasets that were collected with explicit consent for that purpose. Your voice commands never become part of our training data.

We do not create voiceprints or voice profiles that could be used to identify you. We do not analyze the acoustic characteristics of your voice for any purpose beyond the immediate transcription.

We do not retain your voice recordings for quality assurance, debugging, or service improvement. If we need to debug an issue, we work with synthetic test data or data from consenting team members.

We do not share your voice data with any third parties. The audio goes to our transcription system and nowhere else. There are no advertising partners, data brokers, or research organizations receiving your voice data.

We do not build profiles of your voice commands over time. Each voice interaction is processed independently without reference to your previous commands. We do not track what you ask for, what you search for, or what actions you take.

Why We Chose This Approach

You might wonder why we would voluntarily give up access to data that could potentially improve our product. After all, many companies argue that data collection is necessary for providing good service. Our answer is that we believe privacy and quality are not mutually exclusive.

The technical argument that companies need user data to improve speech recognition was more valid in the past when training data was scarce and expensive. Today, there are abundant licensed datasets available for training speech recognition models. We can build an excellent product without treating our users as sources of training data.

More importantly, we believe that privacy should be the default, not a premium feature or a checkbox in settings. When you use Air, you should not have to wonder whether your voice is being recorded, analyzed, or retained. The answer is always no.

This approach also aligns our incentives correctly. We make money by selling software that people find valuable enough to pay for. We do not make money by collecting and monetizing user data. This means our success depends entirely on building a great product, not on harvesting as much personal information as possible.

The Business Case for Privacy

Some people are surprised to learn that a privacy-focused approach can also be a sound business strategy. The reasoning is straightforward.

Trust is increasingly valuable to users. As awareness of data privacy issues grows, users actively seek out products that respect their privacy. By making privacy a core feature of Air, we attract users who care about how their data is handled.

Privacy reduces liability. Companies that accumulate large stores of user data also accumulate legal and regulatory risk. Data breaches are expensive in terms of both direct costs and reputation damage. By not retaining voice data, we eliminate a significant category of risk.

Privacy simplifies compliance. Regulations like GDPR and CCPA impose significant obligations on companies that collect and process personal data. By minimizing data collection, we minimize the compliance burden and the complexity of our privacy infrastructure.

Ultimately, we believe that building products people trust is more sustainable than building products that exploit their data. Air is designed to be a tool that helps you get things done, not a surveillance system disguised as a productivity app.

Technical Safeguards for Voice Privacy

Beyond our policies, we have implemented technical measures that enforce our privacy commitments.

Audio data is encrypted in transit and never written to permanent storage on our servers. The data exists only in memory during the brief processing window and is explicitly cleared afterward.

We do not generate or store any identifiers that could link voice data to specific users. Even if someone gained access to our processing infrastructure, they would find no way to attribute any audio to any individual.

Our logging systems are designed to exclude audio data and detailed transcription content. We log only the information necessary for system monitoring and debugging, and this information does not include the content of your voice commands.

These technical measures ensure that our privacy commitments are enforced at the infrastructure level, not just the policy level. Even if an employee wanted to access user voice data, they would find no mechanism to do so.