diff --git a/docs/reference/class/audio.rst b/docs/reference/class/audio.rst
index 6e86dff..9a74223 100644
--- a/docs/reference/class/audio.rst
+++ b/docs/reference/class/audio.rst
@@ -2,12 +2,17 @@
 Audio
 *****
 
-Role: device only.  TinyUSB supports USB Audio Class 1.0 and 2.0 streaming.
+TinyUSB supports USB Audio Class 1.0 and 2.0 streaming in device and host
+roles.
+
+Device driver
+=============
+
 The descriptors define the topology, formats, channels, rates, controls, and
 alternate settings; the application supplies or consumes the audio samples.
 
 Start from an example
-=====================
+---------------------
 
 Audio descriptors and buffer sizes are tightly coupled.  Copy the closest
 example, confirm that it enumerates, and then change one property at a time:
@@ -20,7 +25,7 @@
   UAC2 at high speed, with multiple rates.
 
 Configuration
-=============
+-------------
 
 Set ``CFG_TUD_AUDIO`` to the number of audio functions.  The principal options
 are:
@@ -71,7 +76,7 @@
 intervals per millisecond than full-speed audio.
 
 Data path
-=========
+---------
 
 .. list-table::
    :header-rows: 1
@@ -112,7 +117,7 @@
 active.
 
 Control requests
-================
+----------------
 
 Implement the ``tud_audio_get_req_*_cb()`` and ``tud_audio_set_req_*_cb()``
 callbacks for every control advertised by the descriptors, such as clock
@@ -129,3 +134,147 @@
 
 Specifications used: *USB Device Class Definition for Audio Devices*, Release
 1.0 and Release 2.0, plus *Audio Data Formats*, Release 2.0.
+
+Host driver
+===========
+
+The host driver discovers one playback stream and one capture stream in each
+supported AudioControl function.  It presents each usable alternate setting
+and discrete sample rate as a complete ``format``, ``sample_rate``, and
+``channels`` configuration.  The application selects one of these tuples;
+TinyUSB manages interface alternate settings, endpoints, packet sizing, and
+explicit feedback.
+
+Start from :doc:`../../examples/host/audio_host`.  It discovers configurations,
+selects signed 16-bit capture and playback streams, and cycles through
+microphone, speaker, and loopback operation.
+
+Configuration
+-------------
+
+Set ``CFG_TUH_AUDIO`` to enable the driver.  The principal options are:
+
+.. list-table::
+   :header-rows: 1
+   :widths: 36 18 46
+
+   * - Option
+     - Default
+     - What it controls
+   * - ``CFG_TUH_AUDIO``
+     - ``0``
+     - Enables the Audio host class.  A nonzero value includes the driver.
+   * - ``CFG_TUH_AUDIO_PROTOCOLS``
+     - UAC1
+     - Bitmask of ``TUH_AUDIO_PROTOCOL_UAC1`` and
+       ``TUH_AUDIO_PROTOCOL_UAC2``.  Combine them to accept both versions.
+   * - ``CFG_TUH_AUDIO_MAX``
+     - ``1``
+     - Maximum number of mounted AudioControl functions.
+   * - ``CFG_TUH_AUDIO_MAX_AS``
+     - ``4``
+     - Maximum number of nonzero-bandwidth alternate settings retained per
+       logical stream.
+   * - ``CFG_TUH_AUDIO_MAX_SAM_FREQ``
+     - ``5``
+     - Maximum discrete sample rates retained per UAC1 alternate setting or
+       UAC2 Clock Source.
+   * - ``CFG_TUH_AUDIO_EPIN_BUFSIZE`` /
+       ``CFG_TUH_AUDIO_EPOUT_BUFSIZE``
+     - ``256`` bytes
+     - Largest capture/playback packet the driver can submit.  A configuration
+       whose packet for one polling interval is larger is rejected.
+   * - ``CFG_TUH_AUDIO_STREAM_BUFSIZE``
+     - ``1024`` bytes
+     - Per-stream FIFO depth.  Playback sends silence when a complete packet
+       is unavailable; capture overwrites the oldest complete frames when the
+       FIFO is full.
+
+Increase the endpoint buffers for high channel counts, sample rates, or sample
+widths.  All buffers are statically allocated, so these maxima directly affect
+RAM use.
+
+Stream lifecycle
+----------------
+
+After ``tuh_audio_mount_cb()``, enumerate stream indices from zero through
+``tuh_audio_stream_count() - 1``.  Use ``tuh_audio_stream_direction()`` to
+distinguish ``TUH_AUDIO_STREAM_PLAYBACK`` (host to device) from
+``TUH_AUDIO_STREAM_CAPTURE`` (device to host).  Then enumerate the stream's
+configurations with ``tuh_audio_config_count()`` and
+``tuh_audio_config_get()``.
+
+For each stream:
+
+1. Call ``tuh_audio_configure()`` with a supported configuration index while
+   the stream is stopped.
+2. Call ``tuh_audio_start()``.  A ``true`` return means that startup was
+   submitted, not that it completed.
+3. Wait for ``TUH_AUDIO_EVENT_START_COMPLETE`` in ``tuh_audio_event_cb()`` and
+   check that its transfer result is successful before handling samples.
+4. Use the non-blocking frame FIFO APIs while the stream runs.
+5. Call ``tuh_audio_stop()`` and wait for
+   ``TUH_AUDIO_EVENT_STOP_COMPLETE`` before reconfiguring the stream.
+
+Capture and playback streams in the same AudioControl function must use the
+same sample rate while they run concurrently.  ``tuh_audio_active_config()``
+returns ``TUSB_INDEX_INVALID_8`` when no configuration is selected.
+
+Audio data is counted in frames, not bytes.  One frame contains one sample for
+every channel; obtain its byte size with ``tuh_audio_config_frame_size()``.
+``tuh_audio_read()`` removes whole frames from a capture FIFO, while
+``tuh_audio_write()`` queues whole frames for playback.  Their return values
+may be shorter than requested.  Use ``tuh_audio_read_available()`` and
+``tuh_audio_write_available()`` to service the FIFOs from the application task;
+the transfer callbacks are notifications and need not drive FIFO servicing.
+
+Callbacks and failures
+----------------------
+
+``tuh_audio_descriptor_cb()`` runs during enumeration, before the mount
+callback.  Its descriptor pointers are valid only during the call.  Copy any
+entity IDs or descriptor fields needed for later raw controls, but do not
+submit control transfers from this callback.
+
+``tuh_audio_capture_cb()`` and ``tuh_audio_playback_cb()`` report successful
+isochronous transfers.  ``tuh_audio_event_cb()`` reports asynchronous start and
+stop completion and ``TUH_AUDIO_EVENT_XFER_FAILED``.  A failed stream has been
+stopped; the application may reconfigure or restart it.  Clear all saved
+indices and associated state in ``tuh_audio_umount_cb()`` because an index may
+be reused by a later device.
+
+Run ``tuh_task()`` continuously.  Isochronous transfers follow the endpoint's
+polling interval, and delaying the host task can exhaust the FIFO even when its
+average producer and consumer rates match.
+
+Feature Unit controls
+---------------------
+
+The driver discovers master mute and volume capabilities before the mount
+callback.  Test mute with ``tuh_audio_mute_supported()`` and retrieve the
+cached volume range with ``tuh_audio_volume_range_get()``.  Volume values use
+signed 1/256 dB units; ``TUH_AUDIO_VOLUME_SILENCE`` represents silence, and
+``TUH_AUDIO_CHANNEL_MASTER`` selects the master channel.
+
+The asynchronous ``tuh_audio_mute_*()`` and ``tuh_audio_volume_*()`` APIs use
+a completion callback.  Their synchronous helpers block until completion and
+should only be used while audio streaming is stopped; synchronous transfers
+can disrupt isochronous traffic.  Use ``tuh_audio_control_xfer()`` for other
+class-specific entity controls.  Buffers passed to asynchronous controls must
+remain valid until their completion callbacks run.
+
+Supported formats and limitations
+---------------------------------
+
+The host accepts Type-I PCM in signed 8-, 16-, packed 24-, 24-in-32-, and
+32-bit little-endian formats.  UAC1 descriptors must list discrete sample
+rates; continuous ranges are not supported.  UAC2 requires a directly
+connected Clock Source.  Clock Selectors, Clock Multipliers, Sampling Rate
+Converters, Clock Validity, and Valid Alternate Settings controls are not
+handled.
+
+Explicit feedback endpoints using 10.14 or 16.16 values are supported.
+Implicit-feedback IN endpoints are treated as ordinary capture endpoints and
+do not pace playback.  ``MaxPacketsOnly`` endpoints are not supported: OUT
+packets are not padded to ``wMaxPacketSize``, and padding in IN packets is not
+removed from the captured data.
diff --git a/docs/reference/class/host.rst b/docs/reference/class/host.rst
index ef42e3d..ebb5a55 100644
--- a/docs/reference/class/host.rst
+++ b/docs/reference/class/host.rst
@@ -2,10 +2,10 @@
 Using Host Classes
 ******************
 
-TinyUSB provides application-level host drivers for CDC serial, HID, MIDI 1.0,
-MIDI 2.0, and Mass Storage.  A host application is asynchronous: a mount
-callback reports a ready interface, I/O is queued, and completion or receive
-callbacks advance the application state.
+TinyUSB provides application-level host drivers for Audio, CDC serial, HID,
+MIDI 1.0, MIDI 2.0, and Mass Storage.  A host application is asynchronous: a
+mount callback reports a ready interface, I/O is queued, and completion or
+receive callbacks advance the application state.
 
 Setup checklist
 ===============
@@ -28,6 +28,7 @@
    #define CFG_TUH_ENABLED    1
    #define CFG_TUH_DEVICE_MAX 4
    #define CFG_TUH_HUB        1
+   #define CFG_TUH_AUDIO      1
    #define CFG_TUH_CDC        1
    #define CFG_TUH_HID        (3 * CFG_TUH_DEVICE_MAX)
    #define CFG_TUH_MSC        1
@@ -137,6 +138,7 @@
 dispatch.  Prefer the asynchronous APIs.  Where a class provides a synchronous
 helper, use it only from a context in which the host task can still run.
 
-Start with :doc:`../../examples/host/cdc_msc_hid` for CDC, HID, and MSC,
+Start with :doc:`../../examples/host/audio_host` for Audio,
+:doc:`../../examples/host/cdc_msc_hid` for CDC, HID, and MSC,
 :doc:`../../examples/host/midi_rx` for MIDI 1.0, or
 :doc:`../../examples/host/midi2_host` for MIDI 2.0.
diff --git a/docs/reference/class/index.rst b/docs/reference/class/index.rst
index 2f8e577..dff38aa 100644
--- a/docs/reference/class/index.rst
+++ b/docs/reference/class/index.rst
@@ -24,7 +24,7 @@
      - Guide
    * - Audio 1.0/2.0
      - Yes
-     - No
+     - Yes
      - :doc:`audio`
    * - Bluetooth HCI
      - Yes