[feat] add input validation, enhanced error messages, and new integration tests

src/lib.rs

@@ -694,10 +694,11 @@ pub fn decode_audio_to_pcm_f32_ffmpeg(audio_path: &Path) -> Result<Vec<f32>> {
         }
     };
     if !output.status.success() {
+        let stderr = String::from_utf8_lossy(&output.stderr);
         return Err(anyhow!(
-            "ffmpeg failed for {}: {}",
+            "Failed to decode audio from {} using ffmpeg. This may indicate the file is not a valid or supported audio/video file, is corrupted, or cannot be opened. ffmpeg stderr: {}",
             audio_path.display(),
-            String::from_utf8_lossy(&output.stderr)
+            stderr.trim()
         ));
     }
     let bytes = output.stdout;

src/main.rs

@@ -186,6 +186,22 @@ fn is_audio_file(path: &Path) -> bool {
     false
 }
 
+fn validate_input_path(path: &Path) -> anyhow::Result<()> {
+    use anyhow::{anyhow, Context};
+    let display = path.display();
+    if !path.exists() {
+        return Err(anyhow!("Input not found: {}", display));
+    }
+    let md = std::fs::metadata(path).with_context(|| format!("Failed to stat input: {}", display))?;
+    if md.is_dir() {
+        return Err(anyhow!("Input is a directory (expected a file): {}", display));
+    }
+    // Attempt to open to catch permission errors early
+    std::fs::File::open(path)
+        .with_context(|| format!("Failed to open input file: {}", display))
+        .map(|_| ())
+}
+
 struct LastModelCleanup {
     path: PathBuf,
 }
@@ -315,6 +331,18 @@ fn run() -> Result<()> {
         return Err(anyhow!("No input files provided"));
     }
 
+    // Preflight: validate each input path and type
+    for inp in &inputs {
+        let p = Path::new(inp);
+        validate_input_path(p)?;
+        if !(is_audio_file(p) || is_json_file(p)) {
+            return Err(anyhow!(
+                "Unsupported input type (expected .json transcript or common audio/video): {}",
+                p.display()
+            ));
+        }
+    }
+
     // Language must be provided via CLI when transcribing audio (no detection from JSON/env)
     let lang_hint: Option<String> = if let Some(ref l) = args.language {
         normalize_lang_code(l).or_else(|| Some(l.trim().to_lowercase()))

src/models.rs

@@ -620,10 +620,13 @@ fn download_one_model(client: &Client, models_dir: &Path, entry: &ModelEntry) ->
         url
     );
     let mut resp = client
-        .get(url)
+        .get(url.clone())
         .send()
         .and_then(|r| r.error_for_status())
-        .context("Failed to download model")?;
+        .with_context(|| format!(
+            "Failed to download model {} from {}. If your terminal has display/TTY issues, try running with --no-progress.",
+            entry.name, url
+        ))?;
 
     let tmp_path = models_dir.join(format!("ggml-{}.bin.part", entry.name));
     if tmp_path.exists() {