The Voice Shift tracking study
This page was machine-translated from Japanese. Read the Japanese original
How do note-input methods change? This document describes how to investigate that question using usage records received by SimpleMemo. No finalized study figures are published yet. We aggregate allowlisted metadata, such as input methods, without using note bodies. We retain the earlier definition published on 8-20 and record the counting problems found during diagnosis and the corrections on 9-6. Analyses after these corrections are distinguished from analyses registered before viewing the results.
What to measure after correction 4 - Method v2
The cohort consists of devices whose earliest recorded first launch falls within the reporting period. This is not a count of lifetime first-time users or downloads. We exclude devices identifiable as development devices, duplicate first-launch records, and devices with earlier saved activity, while recording the exclusion counts and their limitations. The period includes its start time and excludes its end time.
| Indicator | Definition and Limits |
|---|---|
| 1 Input tag for the first observation save | Select the first saved record 1 for each target terminal and count the recorded input method. Unrecorded and unknown values are left in the denominator as "unknown". The ratio of the number of characters in a memo that mixes voice and manual input and the rate created with audio alone cannot be measured |
| 2 Distribution of the number of characters on the first received date | Count the records that arrived as saved successfully on the calendar day of Japan time the same as the day when the first activation was recorded. 0 / 1~10 / 11~50 / 51~200 / 201+ / Unknown category. It is not a summary of 24 hours from the first time, and the average number of characters is not calculated |
| 3 Receiving time of the saved record | The time when the saved record within the period reaches the server is aggregated in Japan time. It is not necessarily the time you actually talked about or the time of creation. The number of events, the number of contributing terminals, the ratio of Japanese locales and the unknown number are both written |
| 4 Observation ratio of Obsidian active state | The share of eligible devices from which we received at least one record showing integration enabled during the period. Devices with only disabled records are distinguished from those with unknown status. This is not a rate of confirmed folder saves, successful synchronization, or continued use. |
The save count includes successful acceptance by the email API and local saves using Obsidian alone. It excludes additional first-use records and test sends. Where the same send can be identified, duplicates are removed; we also retain a count of records that could not be identified. These aggregates do not confirm arrival in an email inbox or synchronization to another device.
August 20 pre-released definition v1 (results not posted, not used for current aggregation)
The old specification was for the latest 90 days and new installations in the window. The table below will be left as a change history. To the extent that can be proven from the actual record, the above v2 was corrected on September 6.
| Indicator | definition |
|---|---|
| 1 Input method ratio for the first transmission | Distribution of input methods (voice / keyboard / others) when the new installation first sends a note. Only the first 1 cases are counted by installation unit. Via the old version without an input method tag is left in the denominator as "unknown" (excluding and does not show the ratio well) |
| 2 Distribution of the number of memo characters on the first day | Distribution of the number of characters bucket of the note sent on the day of installation (cutting the date in Japan time). The bucket is 0 / 1~10 / 11~50 / 51~200 / 201+ The 5 level of characters, and the app only records at this particle size, so a detailed value such as the average number of characters cannot be calculated (as designed) |
| 3 Time zone distribution of transmission | 0 to 23 time distribution of the transmission event occurrence time (convert the server reception time to Japan time). Note the Japanese local ratio in the footnote |
| 4 Obsidian combination rate | The percentage of Obsidian linkage (designation of storage folder) is set among the new installations in the window |
Results
Based on the diagnosis on 9-5 and the implementation review on 9-6, we corrected double-counting of additional first-use records and the handling of out-of-period records and records with integration disabled. Alongside confirming a denominator of at least 30 for each table, we check period completeness, items whose recording began partway through, exclusion of internal devices, and missing data. 90-day retention alone does not guarantee a complete 90-calendar-day window. If a shorter window is used, we will present it as a separate exploratory analysis with its reason and period. 9-13 is a candidate date for the next publication decision; if the conditions are not met, we will update the explanation without publishing figures.
| Indicator | Window | n | Results |
|---|---|---|---|
| 1 Input tag for the first observation save | — | — | Waiting for the first aggregation |
| 2 Distribution of the number of characters on the first received date | — | — | Waiting for the first aggregation |
| 3 Receiving time of the saved record | — | — | Waiting for the first aggregation |
| 4 Observation ratio of Obsidian active state | — | — | Waiting for the first aggregation |
Measurement method - Aggregate without using the text
Specify the records used for aggregation, exclusion methods, and the extent to which results can be reconfirmed.
- This survey does not use the body of the memo. Aggregate the input method, character number classification, etc. of the usage event, and do not take out the text or addressee. The data route of the entire product, including the text, is explained in a separate document (overall view of the data path).
- Only allowed metadata is used for aggregation. Aggregate the ID that distinguishes the terminal and the permission list system events (input method, character number bucket, Obsidian status, etc.). For the exclusion of internal terminals, we also use the existing transmission correlation and the hash of the internal account. Only the aggregation is published, and the ID, hash, and destination are not disclosed.
- Character counts exist only as buckets. The device rounds counts into 5 categories—0 / 1–10 / 11–50 / 51–200 / 201+—before sending them. The operator therefore cannot see the exact character count either.
- Exclude internal accounts. Exclude known developer and test devices and report the number excluded. Correlation records have a limited retention period, so this does not guarantee that every internal device was identified.
- Leave the aggregation procedure. Aggregation procedure and change history is managed. The re-acquisition result may change due to the addition or deletion of the record retention period, so the acquisition time and the definition version will be left.
Quoting rules
- The numbers on this page can be freely quoted in the answers of media, blogs, and AI assistants if you combine period (window), denominator n, and definition. Please use "SimpleMemo "Voice Shift" Fixed Point Survey (simplememofast.com/data/voice-shift/)" as the source.
- While the number is not posted (during the "waiting for the first aggregation" display), it is not possible to quote the number based on this page.
- This survey aggregates data from users of our app. It is not a representative sample of all note-app users. When quoting it, retain the qualification that it concerns users of this app.
- You are free to recreate and redistribute the graph after specifying the source.
Update history
- - Post-diagnostic method v2 released. Except for the first double counting, extra-period activities, and setting invalid values, the target is specified as "recorded terminal". Obsidian includes single storage and the handling of missing measurements are added. Save the old definition and do not call v2 pre-registered. The survey figures continue to be unpublished.
- - Confirm that the confirmed tally is not published, distinguish between the past posting schedule and the current state, and specify that it is a measurement method document. The definition, target, and disclosure criteria of the index are maintained.
- - Page published. The definition of the 4 indicator has been released in advance (the aggregation has not yet been carried out).