Comparison of Speech Recognition use in Android: by Intent or on-thread?

The main difference is UI. SpeechRecognizer doesn't have any so you are responsible for creating one.
I use to wrote a prototype where I've have receiver for listening headset button, then activating speech recognition to listen for some commands. Screen was not activated so I had to use SpeechRecognizer (my UI was some prerecorded sounds and Text To Speech).

Second difference is that SpeechRecognizer has ability for constant listening. Intent version will always end exaction after some period. For example SpeechRecognizer is used by speech recognition "keyboard" so you can dictate a SMS.
In such case you will receive partial results only (in normal mode SpeechRecognizer gives only final results).

One thing that the other answers have not mentioned: if multiple speech recognizers are installed on the device then user switching between them is different depending on if "Intent" or the SpeechRecognizer is used.

• In case of "Intent" the standard Activity selection dialog is popped up. The user can choose the recognizer to be used, and optionally set it globally as the default recognizer, to avoid the dialog in the future.
• In case of SpeechRecognizer the user can set and configure the default recognizer in the global settings (Language and input -> Voice recognizer on ICS).

So, depending on which interface is used the documentation about setting the default recognizer and switching between recognizers should be different. (In most cases though there is just one recognizer, Google Voice Search, so this might not be a big issue in practice.)

In addition to all this, I'd add at least this point:

SpeechRecognizer is better for hands-free user interfaces, since your app actually gets to respond to error conditions like "No matches" and perhaps restart itself. When you use the Intent, the app beeps and shows a dialog that the user must press to continue.

My summary is as follows:

SpeechRecognizer

• Show different UI or no UI at all. Do you really want your app's UI to beep? Do you really want your UI to show a dialog when there is an error and wait for user to click?

• App can do something else while speech recognition is happening

• Can recognize speech while running in the background or from a service

• Can Handle errors better

• Can access low level speech stuff like the raw audio or the RMS. Analyze that audio or use the loudness to make some kind of flashing light to indicate the app is listening

Intent

• Consistent, and easy to use UI for users
• Easy to program