Voicebuds Are Whisper-Level Voice Earbuds for Private AI Use
Voicebuds are whisper-level voice earbuds from Subtle that isolate speech for private dictation, calls, and AI chat in noisy or shared spaces.
Photo source:
voicebuds
Voicebuds Are Whisper-Level Voice Earbuds for Private AI Use
Speaking to a device in an open office usually
means either raising your voice or giving up on voice input altogether. Subtle,
a Palo Alto-based company, built Voicebuds to remove that trade-off. The
earbuds are designed to pick up speech at whisper level and below, letting a
person dictate, take calls, or talk to an AI assistant without being overheard
by the people around them.
Before entering the consumer market with
Voicebuds, Subtle had already worked across the automotive, defense, and
electronics industries, including partnerships with Qualcomm and Nothing,
applying its voice-isolation research to other companies' hardware. Voicebuds
mark the company's first standalone consumer product, moving that same
underlying technology into a device it designed and sells directly. The earbuds
were named a top hardware pick at CES by Engadget following their debut, and
the product is currently shipping in staged waves, with its first batch already
sold out and a second wave scheduled to ship this fall.
Capturing Speech Other Earbuds Can't Hear
Most earbuds are built around listening, using
microphones and processing tuned to capture external sound for noise
cancellation. Voicebuds are built around the opposite problem: capturing the
wearer's own voice as cleanly as possible, even when that voice barely rises
above a whisper. Subtle calls the system behind this the Subtle Voice Engine, a
machine learning model trained end-to-end on real-world noise, motion, and
low-volume speech rather than adapted from a device originally designed for
music playback.
A multi-microphone array positioned for
near-field pickup does the physical capturing, tuned specifically for speech
spoken close to the mouth at low volume rather than sound arriving from a
distance. Calls are processed with on-device isolation, so a person's voice is
filtered locally before it is transmitted, rather than sent raw and cleaned up
elsewhere. Subtle states that this entire chain, from microphone hardware to
firmware to the isolation model itself, was designed from scratch for voice
input rather than retrofitted onto an existing audio product, and the company
describes the resulting architecture as patented.
What a Person Can Actually Do With Them
Once speech is captured, the earbuds connect to
the Subtle app to support several distinct tasks. Dictation works inside any
app on a paired iPhone or Mac, letting a person write by speaking rather than
typing, with the company stating this can be roughly four times faster than
typing by hand. Voice notes can be captured even with the phone in a pocket,
giving a way to record a thought without pulling out a device at all. A
contextual AI chat feature lets a person talk through an idea or ask a question
conversationally, again without needing to speak at a normal volume.
The earbuds also function as ordinary wireless
headphones, handling music playback and phone calls without requiring any of
the voice-AI features to be active. Subtle designed the software to hand off
between devices as well, so a dictation session or notes captured on an iPhone
carry over to a connected Mac without needing to be re-entered. The company
points to specific everyday settings where this matters most: open offices, cafés,
public transit, and moments at home when speaking aloud isn't convenient. Each
pair ships with a charging case and a set of small, medium, and large eartips
for fit, and the earbuds carry an IPX4 water-resistance rating, protecting
against sweat, rain, and splashes during everyday use.
A Different Kind of Benchmark
Rather than comparing Voicebuds only to its own
previous products, Subtle has published a benchmark against several established
earbuds already on the market. According to the company, in its own testing,
Voicebuds produced up to five times fewer transcription errors than Apple
AirPods Pro 3, Samsung Galaxy Buds3, Bose QuietComfort Ultra, and Google Pixel
Buds Pro 2, each paired with standard transcription tools, under noisy or
low-volume speech conditions. Subtle attributes the gap to its purpose-built
voice isolation model rather than the generic speech-recognition systems it
says most earbuds rely on, pointing specifically to lossy audio compression
that it says flattens the voice signal in typical earbuds, along with on-device
latency and processing it describes as incompatible with real-time AI dictation
software.
Privacy is handled as a built-in part of the
design rather than an optional setting. Calls are processed on-device where
possible, dictation and AI chat sessions are encrypted, and an optional privacy
mode automatically deletes audio and transcripts after they are processed. The
product itself does not require a subscription for standard use: music, calls,
and everyday audio work without any membership. AI-driven features come with
one year of the Subtle app included at purchase, after which a person can continue
for a monthly fee or drop to a free tier with a limited number of AI
interactions rather than losing access outright.
Where Voice Interfaces Still Struggle
Voice has been positioned as a faster, more
natural way to interact with software for years, yet typing remains the default
in most shared or noisy settings, largely because speaking aloud in those
environments is either impractical or uncomfortable. Consumer voice assistants
and dictation tools have generally been built and tested in quiet, controlled
conditions, which is part of why their accuracy tends to fall sharply in
real-world settings like open offices or public transit.
Voicebuds represent one attempt to close that
gap directly, but the category itself is still new, and Subtle's benchmark
comparisons come from the company's own testing rather than independent,
third-party verification. Current support is also limited to English, with the
company stating multilingual support will follow. How the hardware performs
across a wide range of accents, languages, and background noise types in
everyday use, beyond the conditions Subtle has tested and published, remains to
be seen as the product reaches more people through its staged shipping waves.
You have exceeded your free limits for viewing our premium content
Please subscribe to have unlimited access to our innovations.
Get The Latest In Innovation
Join Our Newsletter
Where bold ideas find bold readers.
