Sleep Secret
Record the night, classify sounds on-device, get a morning report.
新版本 2.19.4
· 列表卡片显示评分，和详情页是同一个数
· 列表圆环配色跟评分方向一致，高分是绿不是红
· 列表圆环配色跟评分方向一致，高分是绿不是红
Record the night, classify the sounds in it on your phone, and get a report in the morning: when there was sound, what it was, and how long it lasted.
Audio stays on the device by default. There is no account and no server — install it and it works. The only exception is if you configure your own WebDAV storage in the export screen; after that, each night's results are uploaded to a drive under your own name.
Recognition runs on the phone. The model is CED-tiny (5.5M parameters, pretrained on AudioSet), converted to ONNX and run on-device. It does not need a network connection.
What it does
- Night-long sound timeline: when there was sound, what it was, how long it lasted
- Snore index: snoring time as a percentage of analysed time, for comparing one night to another
- Category breakdown and hourly distribution
- Event details: time, category and confidence for each event. Snoring clips can be played back inline, with a waveform and a draggable seek bar
The nine categories it reports: snoring / breathing / cough and throat clearing / sneeze / voice and sleep talking / body movement and bed noise / environmental noise / device noise / silence.
What it does not do
- It is not a medical device and cannot be used to diagnose sleep apnoea or any other condition
- It does not do sleep staging (light / deep / REM). That needs movement or heart rate; audio alone cannot do it
- It does not keep the raw audio of the whole night (8 hours is roughly 920 MB). What is kept is the timing and category of each event, plus any snoring clips you choose to retain
Android only for phones; there is also a Windows build intended for testing. There is no iOS version.
Free software under the GPL-3.0-or-later. Source code: https://github.com/14790897/sleep-secret
Audio stays on the device by default. There is no account and no server — install it and it works. The only exception is if you configure your own WebDAV storage in the export screen; after that, each night's results are uploaded to a drive under your own name.
Recognition runs on the phone. The model is CED-tiny (5.5M parameters, pretrained on AudioSet), converted to ONNX and run on-device. It does not need a network connection.
What it does
- Night-long sound timeline: when there was sound, what it was, how long it lasted
- Snore index: snoring time as a percentage of analysed time, for comparing one night to another
- Category breakdown and hourly distribution
- Event details: time, category and confidence for each event. Snoring clips can be played back inline, with a waveform and a draggable seek bar
The nine categories it reports: snoring / breathing / cough and throat clearing / sneeze / voice and sleep talking / body movement and bed noise / environmental noise / device noise / silence.
What it does not do
- It is not a medical device and cannot be used to diagnose sleep apnoea or any other condition
- It does not do sleep staging (light / deep / REM). That needs movement or heart rate; audio alone cannot do it
- It does not keep the raw audio of the whole night (8 hours is roughly 920 MB). What is kept is the timing and category of each event, plus any snoring clips you choose to retain
Android only for phones; there is also a Windows build intended for testing. There is no iOS version.
Free software under the GPL-3.0-or-later. Source code: https://github.com/14790897/sleep-secret
捐助
版本
雖然在下方可選擇下載 APK 檔案，但要留意這樣的安裝方式將不會收到更新通知，是一種較不安全的下載方法。建議您先安裝 F-Droid 用戶端使用。
下載 F-Droid-
x86_64此版本需要 Android 7.0 或更高的版本。
此套件包由原開發者建置和簽署，並保證與此原始碼 Tarball 保持一致。
-
arm64-v8a此版本需要 Android 7.0 或更高的版本。
此套件包由原開發者建置和簽署，並保證與此原始碼 Tarball 保持一致。
-
armeabi-v7a此版本需要 Android 7.0 或更高的版本。
此套件包由原開發者建置和簽署，並保證與此原始碼 Tarball 保持一致。