diff --git a/gearslip/build.gradle.kts b/gearslip/build.gradle.kts
index 6bf18a7..25f2dfc 100644
--- a/gearslip/build.gradle.kts
+++ b/gearslip/build.gradle.kts
@@ -129,7 +129,6 @@ dependencies {
     implementation(platform(libs.androidx.compose.bom))
     implementation(libs.androidx.compose.ui)
     implementation(libs.androidx.compose.material3)
-    implementation(libs.commons.compress)
     implementation(libs.androidx.compose.material.icons.core)
     implementation(libs.androidx.activity.compose)
     implementation(libs.androidx.lifecycle.runtime.compose)
diff --git a/gearslip/proguard-rules.pro b/gearslip/proguard-rules.pro
index b833500..fcb05fc 100644
--- a/gearslip/proguard-rules.pro
+++ b/gearslip/proguard-rules.pro
@@ -11,10 +11,3 @@
 
 # Speech-to-text: the native library calls back into these classes by name.
 -keep class com.k2fsa.sherpa.onnx.** { *; }
-
-# commons-compress refers to optional codecs Gearslip doesn't ship.
--dontwarn org.tukaani.xz.**
--dontwarn com.github.luben.zstd.**
--dontwarn org.brotli.dec.**
--dontwarn org.objectweb.asm.**
--dontwarn org.osgi.**
diff --git a/gearslip/src/main/java/app/seb3thehacker/gearslip/speech/SpeechModels.kt b/gearslip/src/main/java/app/seb3thehacker/gearslip/speech/SpeechModels.kt
index 328578c..d7cc220 100644
--- a/gearslip/src/main/java/app/seb3thehacker/gearslip/speech/SpeechModels.kt
+++ b/gearslip/src/main/java/app/seb3thehacker/gearslip/speech/SpeechModels.kt
@@ -18,16 +18,14 @@ import java.util.concurrent.Executors
 import kotlinx.coroutines.flow.MutableStateFlow
 import kotlinx.coroutines.flow.StateFlow
 import kotlinx.coroutines.flow.asStateFlow
-import org.apache.commons.compress.archivers.tar.TarArchiveInputStream
-import org.apache.commons.compress.compressors.bzip2.BZip2CompressorInputStream
 
 /**
  * The speech models Gearslip listens with, in two roles: a streaming one that writes the words
  * while the driver talks, and an optional final one that reads the whole phrase again once they
  * stop, for a more accurate result. None ships in the APK: each is fetched once and kept in the
  * app's own storage. sherpa-onnx's author publishes them twice: as .tar.bz2 archives on GitHub and
- * as plain files on Hugging Face. The plain files come first where they exist. Unpacking bzip2 on
- * the phone pinned a core and held a download to 150 KB/s on Wi-Fi that does 19 MB/s.
+ * as plain files on Hugging Face. Gearslip takes the plain files: unpacking bzip2 on the phone
+ * pinned a core and held a download to 150 KB/s on Wi-Fi that does 19 MB/s.
  */
 object SpeechModels {
 
@@ -45,8 +43,6 @@ object SpeechModels {
         /** Empty for Canary, which decodes with attention and has no joiner. */
         val joiner: String,
         val modelType: String = "",
-        /** Whether Hugging Face has the files unpacked; otherwise the archive is unpacked on the phone, slowly. */
-        val plainFiles: Boolean = true,
         /** Where the unpacked files are, when not under the archive's own name on Hugging Face. */
         private val filesAt: String? = null,
     ) {
@@ -75,7 +71,6 @@ object SpeechModels {
         ),
         ;
 
-        val url get() = "https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/$archive.tar.bz2"
         fun fileUrl(name: String) = "https://huggingface.co/${filesAt ?: "csukuangfj/$archive/resolve/main"}/$name"
         val files get() = listOf(encoder, decoder, joiner, "tokens.txt").filter { it.isNotEmpty() }
     }
@@ -215,7 +210,7 @@ object SpeechModels {
         if (model == _selected.value || model == _final.value) SpeechEngine.unload()
     }
 
-    /** Fetches just the files the recognizer reads, plain where it can, else out of the archive. */
+    /** Fetches just the files the recognizer reads. */
     fun download(context: Context, model: Model) {
         val app = context.applicationContext
         val state = states.getValue(model)
@@ -235,30 +230,17 @@ object SpeechModels {
                     state.value = State.Downloading((((done + read) * 100) / total).toInt().coerceIn(0, 99))
                     updateProgress(app)
                 }
-                if (model.plainFiles) {
-                    for (name in model.files) {
-                        val conn = open(model.fileUrl(name))
-                        val length = conn.contentLengthLong
-                        val file = File(partial, name)
-                        CountingStream(conn.inputStream, ::progress).use { input ->
-                            file.outputStream().use { input.copyTo(it, 256 * 1024) }
-                        }
-                        check(length <= 0 || file.length() == length) { "$name arrived cut short" }
-                        done += file.length()
-                    }
-                } else {
-                    val conn = open(model.url)
-                    val counting = CountingStream(conn.inputStream, ::progress)
-                    TarArchiveInputStream(BZip2CompressorInputStream(counting.buffered())).use { tar ->
-                        while (true) {
-                            val entry = tar.nextEntry ?: break
-                            val name = entry.name.substringAfterLast('/')
-                            if (entry.isDirectory || name !in model.files) continue
-                            File(partial, name).outputStream().use { tar.copyTo(it) }
-                        }
+                for (name in model.files) {
+                    val conn = open(model.fileUrl(name))
+                    val length = conn.contentLengthLong
+                    val file = File(partial, name)
+                    CountingStream(conn.inputStream, ::progress).use { input ->
+                        file.outputStream().use { input.copyTo(it, 256 * 1024) }
                     }
+                    check(length <= 0 || file.length() == length) { "$name arrived cut short" }
+                    done += file.length()
                 }
-                check(model.files.all { File(partial, it).isFile }) { "the archive was missing files" }
+                check(model.files.all { File(partial, it).isFile }) { "some files were missing" }
                 target.deleteRecursively()
                 check(partial.renameTo(target)) { "couldn't store the model" }
                 GearslipLog.i("speech: downloaded ${model.label}")
diff --git a/gradle/libs.versions.toml b/gradle/libs.versions.toml
index 0075398..f967a0d 100644
--- a/gradle/libs.versions.toml
+++ b/gradle/libs.versions.toml
@@ -14,7 +14,6 @@ material = "1.14.0"
 constraintlayout = "2.2.1"
 zxing = "3.5.4"
 okhttp = "5.4.0"
-commonsCompress = "1.28.0"
 kotlinxSerializationJson = "1.11.0"
 ossLicenses = "17.5.1"
 webkit = "1.16.0"
@@ -50,7 +49,6 @@ androidx-constraintlayout = { group = "androidx.constraintlayout", name = "const
 google-material = { group = "com.google.android.material", name = "material", version.ref = "material" }
 zxing-core = { group = "com.google.zxing", name = "core", version.ref = "zxing" }
 okhttp = { group = "com.squareup.okhttp3", name = "okhttp", version.ref = "okhttp" }
-commons-compress = { group = "org.apache.commons", name = "commons-compress", version.ref = "commonsCompress" }
 kotlinx-serialization-json = { group = "org.jetbrains.kotlinx", name = "kotlinx-serialization-json", version.ref = "kotlinxSerializationJson" }
 google-oss-licenses = { group = "com.google.android.gms", name = "play-services-oss-licenses", version.ref = "ossLicenses" }
 androidx-webkit = { group = "androidx.webkit", name = "webkit", version.ref = "webkit" }